parallel_alternatives - Man Page

The following features are in some of the comparable tools:

Inputs

I1. Arguments can be read from stdin

I2. Arguments can be read from a file

I3. Arguments can be read from multiple files

I4. Arguments can be read from command line

I5. Arguments can be read from a table

I6. Arguments can be read from the same file using #! (shebang)

I7. Line oriented input as default (Quoting of special chars not needed)

Manipulation of input

M1. Composed command

M2. Multiple arguments can fill up an execution line

M3. Arguments can be put anywhere in the execution line

M4. Multiple arguments can be put anywhere in the execution line

M5. Arguments can be replaced with context

M6. Input can be treated as the complete command line

Outputs

O1. Grouping output so output from different jobs do not mix

O2. Send stderr (standard error) to stderr (standard error)

O3. Send stdout (standard output) to stdout (standard output)

O4. Order of output can be same as order of input

O5. Stdout only contains stdout (standard output) from the command

O6. Stderr only contains stderr (standard error) from the command

O7. Buffering on disk

O8. No temporary files left if killed

O9. Test if disk runs full during run

O10. Output of a line bigger than 4 GB

Execution

E1. Run jobs in parallel

E2. List running jobs

E3. Finish running jobs, but do not start new jobs

E4. Number of running jobs can depend on number of cpus

E5. Finish running jobs, but do not start new jobs after first failure

E6. Number of running jobs can be adjusted while running

E7. Only spawn new jobs if load is less than a limit

E8. Full command has non-zero exit value if one job has non-zero exit value

E9. Jobs can be started without reading all input first

Remote execution

R1. Jobs can be run on remote computers

R2. Basefiles can be transferred

R3. Argument files can be transferred

R4. Result files can be transferred

R5. Cleanup of transferred files

R6. No config files needed

R7. Do not run more than SSHD's MaxStartups can handle

R8. Configurable SSH command

R9. Retry if connection breaks occasionally

Semaphore

S1. Possibility to work as a mutex

S2. Possibility to work as a counting semaphore

Legend

- = no

x = not applicable

ID = yes

As every new version of the programs are not tested the table may be outdated. Please file a bug report if you find errors (See REPORTING BUGS).

parallel:

I1 I2 I3 I4 I5 I6 I7

M1 M2 M3 M4 M5 M6

O1 O2 O3 O4 O5 O6 O7 O8 O9 O10

E1 E2 E3 E4 E5 E6 E7 E8 E9

R1 R2 R3 R4 R5 R6 R7 R8 R9

S1 S2

Summary (see legend above):

I1 I2 - - - - -

- M2 M3 - - -

- O2 O3 - O5 O6

E1 - - - - - - E8 E9

- - - - - x - - -

- -

xargs offers some of the same possibilities as GNU parallel.

xargs deals badly with special characters (such as space, \, ' and "). To see the problem try this:

  touch important_file
  touch 'not important_file'
  ls not* | xargs rm
  mkdir -p "My brother's 12\" records"
  ls | xargs rmdir
  touch 'c:\windows\system32\clfs.sys'
  echo 'c:\windows\system32\clfs.sys' | xargs ls -l

You can specify -0, but many input generators are not optimized for using NUL as separator but are optimized for newline as separator. E.g. awk, ls, echo, tar -v, head (requires using -z), tail (requires using -z), sed (requires using -z), perl (-0 and \0 instead of \n), locate (requires using -0), find (requires using -print0), grep (requires using -z or -Z), sort (requires using -z).

GNU parallel's newline separation can be emulated with:

  cat | xargs -d "\n" -n1 command

xargs can run a given number of jobs in parallel, but has no support for running number-of-cpu-cores jobs in parallel.

xargs has no support for grouping the output, therefore output may run together, e.g. the first half of a line is from one process and the last half of the line is from another process. The example Parallel grep cannot be done reliably with xargs because of this. To see this in action try:

  parallel perl -e "'"'$a="1"."{}"x100000000;print $a,"\n"'"'" \
    '>' {} ::: a b c d e f g h
  # Serial = no mixing = the wanted result
  # 'tr -s a-z' squeezes repeating letters into a single letter
  echo a b c d e f g h | xargs -P1 -n1 grep 1 | tr -s a-z
  # Compare to 8 jobs in parallel
  parallel -kP8 -n1 grep 1 ::: a b c d e f g h | tr -s a-z
  echo a b c d e f g h | xargs -P8 -n1 grep 1 | tr -s a-z
  echo a b c d e f g h | xargs -P8 -n1 grep --line-buffered 1 | \
    tr -s a-z

Or try this:

  slow_seq() {
    echo Count to "$@"
    seq "$@" |
      perl -ne '$|=1; for(split//){ print; select($a,$a,$a,0.100);}'
  }
  export -f slow_seq
  # Serial = no mixing = the wanted result
  seq 8 | xargs -n1 -P1 -I {} bash -c 'slow_seq {}'
  # Compare to 8 jobs in parallel
  seq 8 | parallel -P8 slow_seq {}
  seq 8 | xargs -n1 -P8 -I {} bash -c 'slow_seq {}'

xargs has no support for keeping the order of the output, therefore if running jobs in parallel using xargs the output of the second job cannot be postponed till the first job is done.

xargs has no support for running jobs on remote computers.

xargs has no support for context replace, so you will have to create the arguments.

If you use a replace string in xargs (-I) you can not force xargs to use more than one argument.

Quoting in xargs works like -q in GNU parallel. This means composed commands and redirection require using bash -c.

  ls | parallel "wc {} >{}.wc"
  ls | parallel "echo {}; ls {}|wc"

becomes (assuming you have 8 cores and that none of the filenames contain space, " or ').

  ls | xargs -d "\n" -P8 -I {} bash -c "wc {} >{}.wc"
  ls | xargs -d "\n" -P8 -I {} bash -c "echo {}; ls {}|wc"

A more extreme example can be found on: https://unix.stackexchange.com/q/405552/

https://www.gnu.org/software/findutils/

Summary (see legend above):

-  -  -  x  -  x  -

-  M2 M3 -  -  -  -

-  O2 O3 O4 O5 O6

-  -  -  -  -  -  -

-  -  -  -  -  -  -  -  -

x  x

find -exec offers some of the same possibilities as GNU parallel.

find -exec only works on files. Processing other input (such as hosts or URLs) will require creating these inputs as files. find -exec has no support for running commands in parallel.

https://www.gnu.org/software/findutils/ (Last checked: 2019-01)

Summary (see legend above):

-  -  -  -  -  -  -

-  -  -  -  -  -

O1 O2 O3 -  x  O6

E1 -  -  -  E5 -

-  -  -  -  -  -  -  -  -

-  -

make -j can run jobs in parallel, but requires a crafted Makefile to do this. That results in extra quoting to get filenames containing newlines to work correctly.

make -j computes a dependency graph before running jobs. Jobs run by GNU parallel does not depend on each other.

(Very early versions of GNU parallel were coincidentally implemented using make -j).

https://www.gnu.org/software/make/ (Last checked: 2019-01)

Summary (see legend above):

I1 I2 - - - - I7

M1 - M3 - - M6

O1 - - x - -

E1 E2 ?E3 E4 - - -

R1 R2 R3 R4 - - ?R7 ? ?

- -

ppss is also a tool for running jobs in parallel.

The output of ppss is status information and thus not useful for using as input for another command. The output from the jobs are put into files.

The argument replace string ($ITEM) cannot be changed. Arguments must be quoted - thus arguments containing special characters (space '"&!*) may cause problems. More than one argument is not supported. Filenames containing newlines are not processed correctly. When reading input from a file null cannot be used as a terminator. ppss needs to read the whole input file before starting any jobs.

Output and status information is stored in ppss_dir and thus requires cleanup when completed. If the dir is not removed before running ppss again it may cause nothing to happen as ppss thinks the task is already done. GNU parallel will normally not need cleaning up if running locally and will only need cleaning up if stopped abnormally and running remote (--cleanup may not complete if stopped abnormally). The example Parallel grep would require extra postprocessing if written using ppss.

For remote systems PPSS requires 3 steps: config, deploy, and start. GNU parallel only requires one step.

EXAMPLES FROM ppss MANUAL

Here are the examples from ppss's manual page with the equivalent using GNU parallel:

  1$ ./ppss.sh standalone -d /path/to/files -c 'gzip '

  1$ find /path/to/files -type f | parallel gzip

  2$ ./ppss.sh standalone -d /path/to/files \
       -c 'cp "$ITEM" /destination/dir '

  2$ find /path/to/files -type f | parallel cp {} /destination/dir

  3$ ./ppss.sh standalone -f list-of-urls.txt -c 'wget -q '

  3$ parallel -a list-of-urls.txt wget -q

  4$ ./ppss.sh standalone -f list-of-urls.txt -c 'wget -q "$ITEM"'

  4$ parallel -a list-of-urls.txt wget -q {}

  5$ ./ppss config -C config.cfg -c 'encode.sh ' -d /source/dir \
       -m 192.168.1.100 -u ppss -k ppss-key.key -S ./encode.sh \
       -n nodes.txt -o /some/output/dir --upload --download;
     ./ppss deploy -C config.cfg
     ./ppss start -C config

  5$ # parallel does not use configs. If you want
     # a different username put it in nodes.txt: user@hostname
     find source/dir -type f |
       parallel --sshloginfile nodes.txt --trc {.}.mp3 \
         lame -a {} -o {.}.mp3 --preset standard --quiet

  6$ ./ppss stop -C config.cfg

  6$ killall -TERM parallel

  7$ ./ppss pause -C config.cfg

  7$ Press: CTRL-Z or killall -SIGTSTP parallel

  8$ ./ppss continue -C config.cfg

  8$ Enter: fg or killall -SIGCONT parallel

  9$ ./ppss.sh status -C config.cfg

  9$ killall -SIGUSR2 parallel

https://github.com/louwrentius/PPSS (Last checked: 2010-12)

Summary (see legend above):

I1 I2 - I4 I5 - -

M1 - M3 - - M6

O1 O2 O3 - O5 O6

E1 - - E4 - E6 -

R1 - - - - R6 - - -

S1 -

pexec is also a tool for running jobs in parallel.

EXAMPLES FROM pexec MANUAL

Here are the examples from pexec's info page with the equivalent using GNU parallel:

  1$ pexec -o sqrt-%s.dat -p "$(seq 10)" -e NUM -n 4 -c -- \
       'echo "scale=10000;sqrt($NUM)" | bc'

  1$ seq 10 | parallel -j4 'echo "scale=10000;sqrt({})" | \
       bc > sqrt-{}.dat'

  2$ pexec -p "$(ls myfiles*.ext)" -i %s -o %s.sort -- sort

  2$ ls myfiles*.ext | parallel sort {} ">{}.sort"

  3$ pexec -f image.list -n auto -e B -u star.log -c -- \
       'fistar $B.fits -f 100 -F id,x,y,flux -o $B.star'

  3$ parallel -a image.list \
       'fistar {}.fits -f 100 -F id,x,y,flux -o {}.star' 2>star.log

  4$ pexec -r *.png -e IMG -c -o - -- \
       'convert $IMG ${IMG%.png}.jpeg ; "echo $IMG: done"'

  4$ ls *.png | parallel 'convert {} {.}.jpeg; echo {}: done'

  5$ pexec -r *.png -i %s -o %s.jpg -c 'pngtopnm | pnmtojpeg'

  5$ ls *.png | parallel 'pngtopnm < {} | pnmtojpeg > {}.jpg'

  6$ for p in *.png ; do echo ${p%.png} ; done | \
       pexec -f - -i %s.png -o %s.jpg -c 'pngtopnm | pnmtojpeg'

  6$ ls *.png | parallel 'pngtopnm < {} | pnmtojpeg > {.}.jpg'

  7$ LIST=$(for p in *.png ; do echo ${p%.png} ; done)
     pexec -r $LIST -i %s.png -o %s.jpg -c 'pngtopnm | pnmtojpeg'

  7$ ls *.png | parallel 'pngtopnm < {} | pnmtojpeg > {.}.jpg'

  8$ pexec -n 8 -r *.jpg -y unix -e IMG -c \
       'pexec -j -m blockread -d $IMG | \
        jpegtopnm | pnmscale 0.5 | pnmtojpeg | \
        pexec -j -m blockwrite -s th_$IMG'

  8$ # Combining GNU B<parallel> and GNU B<sem>.
     ls *jpg | parallel -j8 'sem --id blockread cat {} | jpegtopnm |' \
       'pnmscale 0.5 | pnmtojpeg | sem --id blockwrite cat > th_{}'

     # If reading and writing is done to the same disk, this may be
     # faster as only one process will be either reading or writing:
     ls *jpg | parallel -j8 'sem --id diskio cat {} | jpegtopnm |' \
       'pnmscale 0.5 | pnmtojpeg | sem --id diskio cat > th_{}'

https://www.gnu.org/software/pexec/ (Last checked: 2024-06)

xjobs is also a tool for running jobs in parallel. It only supports running jobs on your local computer.

xjobs deals badly with special characters just like xargs. See the section DIFFERENCES BETWEEN xargs AND GNU Parallel.

EXAMPLES FROM xjobs MANUAL

Here are the examples from xjobs's man page with the equivalent using GNU parallel:

  1$ ls -1 *.zip | xjobs unzip

  1$ ls *.zip | parallel unzip

  2$ ls -1 *.zip | xjobs -n unzip

  2$ ls *.zip | parallel unzip >/dev/null

  3$ find . -name '*.bak' | xjobs gzip

  3$ find . -name '*.bak' | parallel gzip

  4$ ls -1 *.jar | sed 's/\(.*\)/\1 > \1.idx/' | xjobs jar tf

  4$ ls *.jar | parallel jar tf {} '>' {}.idx

  5$ xjobs -s script

  5$ cat script | parallel

  6$ mkfifo /var/run/my_named_pipe;
     xjobs -s /var/run/my_named_pipe &
     echo unzip 1.zip >> /var/run/my_named_pipe;
     echo tar cf /backup/myhome.tar /home/me >> /var/run/my_named_pipe

  6$ mkfifo /var/run/my_named_pipe;
     cat /var/run/my_named_pipe | parallel &
     echo unzip 1.zip >> /var/run/my_named_pipe;
     echo tar cf /backup/myhome.tar /home/me >> /var/run/my_named_pipe

https://www.maier-komor.de/xjobs.html (Last checked: 2019-01)

prll is also a tool for running jobs in parallel. It does not support running jobs on remote computers.

prll encourages using BASH aliases and BASH functions instead of scripts. GNU parallel supports scripts directly, functions if they are exported using export -f, and aliases if using env_parallel.

prll generates a lot of status information on stderr (standard error) which makes it harder to use the stderr (standard error) output of the job directly as input for another program.

EXAMPLES FROM prll's MANUAL

Here is the example from prll's man page with the equivalent using GNU parallel:

  1$ prll -s 'mogrify -flip $1' *.jpg

  1$ parallel mogrify -flip ::: *.jpg

https://github.com/exzombie/prll (Last checked: 2024-06)

dxargs is also a tool for running jobs in parallel.

dxargs does not deal well with more simultaneous jobs than SSHD's MaxStartups. dxargs is only built for remote run jobs, but does not support transferring of files.

https://web.archive.org/web/20120518070250/http://www. semicomplete.com/blog/geekery/distributed-xargs.html (Last checked: 2019-01)

middleman(mdm) is also a tool for running jobs in parallel.

EXAMPLES FROM middleman's WEBSITE

Here are the shellscripts of https://web.archive.org/web/20110728064735/http://mdm. berlios.de/usage.html ported to GNU parallel:

  1$ seq 19 | parallel buffon -o - | sort -n > result
     cat files | parallel cmd
     find dir -execdir sem cmd {} \;

https://github.com/cklin/mdm (Last checked: 2014-06)

xapply can run jobs in parallel on the local computer.

EXAMPLES FROM xapply's MANUAL

Here are the examples from xapply's man page with the equivalent using GNU parallel:

  1$ xapply '(cd %1 && make all)' */

  1$ parallel 'cd {} && make all' ::: */

  2$ xapply -f 'diff %1 ../version5/%1' manifest | more

  2$ parallel diff {} ../version5/{} < manifest | more

  3$ xapply -p/dev/null -f 'diff %1 %2' manifest1 checklist1

  3$ parallel --link diff {1} {2} :::: manifest1 checklist1

  4$ xapply 'indent' *.c

  4$ parallel indent ::: *.c

  5$ find ~ksb/bin -type f ! -perm -111 -print | \
       xapply -f -v 'chmod a+x' -

  5$ find ~ksb/bin -type f ! -perm -111 -print | \
       parallel -v chmod a+x

  6$ find */ -... | fmt 960 1024 | xapply -f -i /dev/tty 'vi' -

  6$ sh <(find */ -... | parallel -s 1024 echo vi)

  6$ find */ -... | parallel -s 1024 -Xuj1 vi

  7$ find ... | xapply -f -5 -i /dev/tty 'vi' - - - - -

  7$ sh <(find ... | parallel -n5 echo vi)

  7$ find ... | parallel -n5 -uj1 vi

  8$ xapply -fn "" /etc/passwd

  8$ parallel -k echo < /etc/passwd

  9$ tr ':' '\012' < /etc/passwd | \
       xapply -7 -nf 'chown %1 %6' - - - - - - -

  9$ tr ':' '\012' < /etc/passwd | parallel -N7 chown {1} {6}

  10$ xapply '[ -d %1/RCS ] || echo %1' */

  10$ parallel '[ -d {}/RCS ] || echo {}' ::: */

  11$ xapply -f '[ -f %1 ] && echo %1' List | ...

  11$ parallel '[ -f {} ] && echo {}' < List | ...

https://www.databits.net/~ksb/msrc/local/bin/xapply/xapply.html (Last checked: 2010-12)

apply can build command lines based on a template and arguments - very much like GNU parallel. apply does not run jobs in parallel. apply does not use an argument separator (like :::); instead the template must be the first argument.

EXAMPLES FROM IBM's KNOWLEDGE CENTER

Here are the examples from IBM's Knowledge Center and the corresponding command using GNU parallel:

To obtain results similar to those of the ls command, enter:

  1$ apply echo *
  1$ parallel echo ::: *

To compare the file named a1 to the file named b1, and the file named a2 to the file named b2, enter:

  2$ apply -2 cmp a1 b1 a2 b2
  2$ parallel -N2 cmp ::: a1 b1 a2 b2

To run the who command five times, enter:

  3$ apply -0 who 1 2 3 4 5
  3$ parallel -N0 who ::: 1 2 3 4 5

To link all files in the current directory to the directory /usr/joe, enter:

  4$ apply 'ln %1 /usr/joe' *
  4$ parallel ln {} /usr/joe ::: *

https://www-01.ibm.com/support/knowledgecenter/ ssw_aix_71/com.ibm.aix.cmds1/apply.htm (Last checked: 2019-01)

Summary (see legend above):

I1 - - - - - I7

- - M3 - - -

(O1) O2 O3 (O4) (O5) O6 - O8 x -

E1 - - - (E5) - - -

R1 - - - x R6 - R8 R9

- -

paexec can run jobs in parallel on both the local and remote computers.

paexec has a job dependency facility so a job can depend on another job to be executed successfully. Sort of a poor-man's make. This can partly be emulated in GNU parallel with tsort.

paexec fails if output of a single line is > 2 GB. Output of a 2 GB line requires 6 GB RAM. Lines of standard output is interleaved (but there is no half line mixing), and output of standard error mixes. Combined with paexec_reorder output order can be the same as input order. In certain situations paexec will eat the last newline of standard output.

There seems to be no way to have the number og jobs depend on the number of CPU threads in a mixed server setup: E.g run 4 jobs on a remote server with 4 cores and 16 jobs on a remote server with 16 cores.

EXAMPLES FROM man paexec

Here are the examples from man paexec with the equivalent using GNU parallel.

  1$ paexec -t '/usr/bin/ssh -x' -n 'host1 host2 host3' \
       -le -g -c calculate-me < tasks.txt |
     paexec_reorder -Mf -Sl

  # GNU Parallel cannot stop processing jobs that depend on another.
  # It can either try all:
  1$ tsort tasks.txt |
       parallel --ssh '/usr/bin/ssh -x' -S "host1,host2,host3" \
         --tagstring {#} --pipe -N1 --log my.log calculate-me

  # Or it can stop at the first failing:
  1$ tsort tasks.txt |
       parallel --ssh '/usr/bin/ssh -x' -S "host1,host2,host3" \
         --tagstring {#} --halt now,fail=1 --pipe -N1 --log my.log calculate-me

  # To retry the the failed and missing tasks:
  1$ tsort tasks.txt |
       parallel --ssh '/usr/bin/ssh -x' -S "host1,host2,host3" \
         --tagstring {#} --halt now,fail=1 --pipe -N1 --joblog my.log \
         --resume-failed calculate-me

  2$ ls -1 *.wav | paexec -x -n +4 -c 'oggenc -Q'

  2$ ls -1 *.wav | parallel -j4 oggenc -Q

  3$ ls -1 *.wav | paexec -xCil -n+4 flac -f --silent

  3$ ls -1 *.wav | parallel --tagstring {#} -j4 'echo {}; flac -f --silent {}'

  4$ { uname -s; uname -r; uname -m; } |
       paexec -x -lp -n+2 -c banner |
       paexec_reorder -l

  4$ { uname -s; uname -r; uname -m; } |
       parallel --tagstring '{#}' -k \
         'banner {} | perl -pe "s/^/getppid().\" \"/e"'

  5$ find . -name '*.dat' -print0 |
       paexec -0 -n+10 -C -J// scp // remoteserver:/remote/path

  5$ find . -name '*.dat' -print0 |
       parallel -0 -j10 -I// scp // remoteserver:/remote/path

  6$ ls -1 *.txt | paexec -n+10 -J%% -c 'awk "BEGIN {print toupper(\"%%\")}"'

  6$ ls -1 *.txt | parallel -j10 -I%% 'awk "BEGIN {print toupper(\"%%\")}"'

EXAMPLES FROM presentation/paexec.tex

  7$ ls -1 *.wav | \
     paexec -x -c 'flac -s' -n +4 > /dev/null

  7$ ls -1 *.wav | \
     parallel -j4 flac -s > /dev/null

  8$ cat ~/bin/toupper
     #!/usr/bin/awk -f
     {
        print " ", toupper($0)
        print ""  # empty line -- end-of-task marker!
        fflush()  # We must flush stdout!
     }

     cat tasks
     apple
     bananas
     orange

     paexec -t ssh -c ~/bin/toupper -n 'server1 server2' < tasks

  8$ parallel --pipe -n1 -S server1,server2 ~/bin/toupper < tasks

  9$ paexec -lr -t ssh -c ~/bin/toupper -n 'server1 server2' < tasks

  9$ # GNU Parallel has no easy way to prepend the server
     parallel --tagstring {#} --pipe -n1 -S server1,server2 ~/bin/toupper < tasks
     cat tasks | parallel --tagstring {#} --pipe -n1 -S server1,server2 --plus \
       ~/bin/toupper '| perl -pe "s/^/{sshlogin}/"'  < tasks

  10$ paexec -n +4 -c ~/bin/toupper < tasks

  10$ parallel -j4 --pipe -n1 ~/bin/toupper < tasks

  11$ paexec -x -t ssh -n 'server1 server2' \
        -c "awk 'BEGIN {print toupper(ARGV[1])}' "  < tasks

  11$ parallel  -S 'server1,server2' \
        "awk 'BEGIN {print toupper(ARGV[1])}'" < tasks

  12$ paexec -x -C -t ssh -n 'server1 server2' \
        awk 'BEGIN {print toupper(ARGV[1])}' < tasks

  12$ parallel -S 'server1,server2' -q \
        awk 'BEGIN {print toupper(ARGV[1])}' < tasks

  13$ paexec -Z240 -x -t ssh -n 'server1 badhostname server2' \
        -c "awk 'BEGIN {print toupper(ARGV[1])}' " < tasks

  13$ parallel --filter-hosts -S 'server1,badhostname,server2' \
        "awk 'BEGIN {print toupper(ARGV[1])}' " < tasks

  14$ cat ~/bin/pbanner
      #!/usr/bin/env sh
      while read task; do
         banner -f M "$task" | pv -qL 300
         echo "$PAEXEC_EOT" # end-of-task marker
      done

      cat tasks
      pae
      xec

      paexec -l -mt='SE@X-L0S0!&' -c ~/bin/pbanner -n +2 < tasks |
        paexec_reorder -mt='SE@X-L0S0!&'

  14$ paexec -y -lc ~/bin/pbanner -n+2 < tasks | paexec_reorder -y

  14$ paexec -l -x -c banner -n+2 < tasks

  14$ parallel --pipe -n1 -j2 ~/bin/pbanner < tasks

  16$ cat ~/tmp/packages_to_build
      audio/cd-discid audio/abcde
      textproc/gsed audio/abcde
      audio/cdparanoia audio/abcde
      audio/id3v2 audio/abcde
      audio/id3 audio/abcde
      misc/mkcue audio/abcde
      shells/bash audio/abcde
      devel/libtool-base audio/cdparanoia
      devel/gmake audio/cdparanoia
      devel/libtool-base audio/id3lib
      devel/gmake audio/id3v2
      audio/id3lib audio/id3v2
      devel/m4 devel/bison
      lang/f2c devel/libtool-base
      devel/gmake misc/mkcue
      devel/bison shells/bash

      cat ~/bin/pkg_builder
      #!/usr/bin/awk -f
      
      {
         print "build " $0
         print "success" # build succeeded! (paexec -ms=)
         print ""        # end-of-task marker
         fflush()        # we must flush stdout
      }

      paexec -g -l -c ~/bin/pkg_builder -n 'server2 server1' \
        -t ssh < ~/tmp/packages_to_build | paexec_reorder

  # GNU Parallel cannot postpone jobs that depend on another.
  # In some cases this will work
  16$ tsort ~/tmp/packages_to_build | parallel -S server2,server1 \
        --pipe -n1 ~/bin/pkg_builder

  17$ cat ~/bin/pkg_builder
      #!/usr/bin/awk -f
      
      {
         print "build " $0
         if ($0 == "devel/gmake")
            print "failure" # Oh no...
            exit 255     # Exit value needed for GNU Parallel
         else
            print "success" # build succeeded!
      
         print ""        # end-of-task marker
         fflush()        # we must flush stdout
      }

      paexec -gl -c ~/bin/pkg_builder -n 'server2 server1' \
        -t ssh < ~/tmp/packages_to_build | paexec_reorder

  # GNU Parallel cannot refrain from starting jobs, that depend on others
  # In some cases this will work
  17$ tsort ~/tmp/packages_to_build | parallel -S server2,server1 \
        --halt now,fail=1 --pipe -n1 ~/bin/pkg_builder

  18$ cat ~/bin/pkg_builder
      #!/usr/bin/awk -f
      
      {
         "hostname -s" | getline hostname
         print "build " $0 " on " hostname
      
         if (hostname == "server1" && $0 == "textproc/gsed")
            exit 139
            # Damn it, I'm dying...
            # Exit value is needed by GNU Parallel
         else
            print "success" # Yes! :-)
      
         print ""        # end-of-task marker
         fflush()        # we must flush stdout
      }
      
      paexec -gl -Z300 -t ssh -c ~/bin/pkg_builder \
        -n 'server2 server1' < ~/tmp/packages_to_build |
        paexec_reorder > result

  # GNU Parallel retries a job on another server, if --retries > 1
  17$ tsort ~/tmp/packages_to_build | parallel -S server2,server1 \
        --halt now,fail=1 --retries 2 --pipe -n1 ~/bin/pkg_builder

  18$ ls -1 *.wav | paexec -x -c 'flac -s' -n+3 >/dev/null

  18$ ls -1 *.wav | parallel -j3 flac -s >/dev/null

  19$ ls -1 *.wav | paexec -ixC -n+3 oggenc -Q | grep .

  19$ ls -1 *.wav | parallel -j3 'echo {}; oggenc -Q {}' | grep .

  20$ cat calc
      #!/bin/sh
      # $1 -- task given on input
      if test $1 = huge; then
          sleep 6
      else
          sleep 1
      fi
      echo "task $1 done"

      printf 'small1\nsmall2\nsmall3\nsmall4\nsmall5\nhuge\n' |
        time -p paexec -c ~/bin/calc -n +2 -xg | grep -v success

  20$ printf 'small1\nsmall2\nsmall3\nsmall4\nsmall5\nhuge\n' |
        time -p parallel -j2 ~/bin/calc | grep -v success

  21$ printf 'small1\nsmall2\nsmall3\nsmall4\nweight: huge 6\n' |
        time -p paexec -c ~/bin/calc -n +2 -x -W1 | grep -v success

  21$ # GNU Parallel does not support weighted jobs.
      # It can be simulated by sorting:
      printf 'small1\nsmall2\nsmall3\nsmall4\nweight: huge 6\n' |
        perl -pe 's/^weight: (.*) (\d+)/$2 $1/ or s/^/1 /' |
        sort -nr | time parallel ~/bin/calc '{=s/^\d* //=}' |
        grep -v success

EXAMPLES FROM paexec's example dir

Here are the examples from paexec's example dir with the equivalent using GNU parallel:

all_substr

  $ paexec -lpe -c "`pwd`/cmd" -n +3 <<EOF

  $ parallel -j3 --pipe -n1 --tagstring {#} \
      './cmd | perl -pe "s/^/getppid().\" \"/e"' <<EOF

cc_wrapper

  $ paexec -c "env CC=gcc CFLAGS=-O2 `pwd`/cmd" \
             -n 'host1 host2' \
             -t '/usr/bin/ssh -x' <<EOF

  $ parallel --pipe -n1 -S 'host1,host2' \
      "env CC=gcc CFLAGS=-O2 `pwd`/cmd" <<EOF

  # This is not exactly the same, but avoids the wrapper
  $ parallel -S host1,host2 gcc -O2 -c -o {.}.o {} <<EOF

cc_wrapper2

  $ ls -1 $PWD/*.c | paexec -c "env $CC $CFLAGS -c " -n +4 -x

  $ ls -1 $PWD/*.c | parallel -j4 "env $CC $CFLAGS -c"

dirtest

  $ paexec -gx -l -c 'test -d' -md=';' -n +3 < tasks

  # GNU Parallel cannot refrain from starting jobs, that depend on others
  $ parallel -j3 --tag test -d '{= s/.*;// =}; echo $?'  < tasks

divide

  $ paexec -s -l -c cmd_divide -n +3 <<EOF

  # GNU Parallel cannot refrain from starting jobs, that depend on others
  $ parallel -j3 --pipe -n1 cmd_divide <<EOF

make_package

  1$ paexec -g -le -c "`pwd`/cmd" -n +3 < tasks | paexec_reorder -g -Ms

  # GNU Parallel cannot refrain from starting jobs, that depend on others
  1$ cat < tasks | parallel --pipe -n1 -j3 "`pwd`/cmd"

  2$ paexec -g -le -c "`pwd`/cmd" -n +3 < tasks_cycle

  2$ tsort < tasks_cycle | parallel --pipe -n1 -j3 "`pwd`/cmd"

toupper

  $ input | paexec -c "`pwd`/cmd" -n +2 | cut -b 2-

  $ input | parallel --pipe -n1 -j2 "`pwd`/cmd" | cut -b 2-

  $ # Without the wrapper:
    input | parallel --pipe -n1 -j2 'awk {print\ toupper\(\$0\)}'

wav2flac

  $ ls -1 "$dir"/*.wav | paexec -x -c 'flac --silent' -n +"$num"

  $ ls -1 "$dir"/*.wav | parallel -j $num flac --silent

https://github.com/cheusov/paexec (Last checked: 2024-06)

Summary (see legend above):

I1 - - I4 - - (I7)

M1 (M2) M3 (M4) M5 M6

- O2 O3 - O5 - - x x O10

E1 - - - - - -

- - - - - - - - -

- -

(I7): Only under special circumstances. See below.

(M2+M4): Only if there is a single replacement string.

map rejects input with special characters:

  echo "The Cure" > My\ brother\'s\ 12\"\ records

  ls | map 'echo %; wc %'

It works with GNU parallel:

  ls | parallel 'echo {}; wc {}'

Under some circumstances it also works with map:

  ls | map 'echo % works %'

But tiny changes make it reject the input with special characters:

  ls | map 'echo % does not work "%"'

This means that many UTF-8 characters will be rejected. This is by design. From the web page: "As such, programs that quietly handle them, with no warnings at all, are doing their users a disservice."

map delays each job by 0.01 s. This can be emulated by using parallel --delay 0.01.

map prints '+' on stderr when a job starts, and '-' when a job finishes. This cannot be disabled. parallel has --bar if you need to see progress.

map's replacement strings (% %D %B %E) can be simulated in GNU parallel by putting this in ~/.parallel/config:

  --rpl '%'
  --rpl '%D $_=Q(::dirname($_));'
  --rpl '%B s:.*/::;s:\.[^/.]+$::;'
  --rpl '%E s:.*\.::'

map does not have an argument separator on the command line, but uses the first argument as command. This makes quoting harder which again may affect readability. Compare:

  map -p 2 'perl -ne '"'"'/^\S+\s+\S+$/ and print $ARGV,"\n"'"'" *

  parallel -q perl -ne '/^\S+\s+\S+$/ and print $ARGV,"\n"' ::: *

map can do multiple arguments with context replace, but not without context replace:

  parallel --xargs echo 'BEGIN{'{}'}END' ::: 1 2 3

  map "echo 'BEGIN{'%'}END'" 1 2 3

map has no support for grouping. So this gives the wrong results:

  parallel perl -e '\$a=\"1{}\"x100000000\;print\ \$a,\"\\n\"' '>' {} \
    ::: a b c d e f
  ls -l a b c d e f
  parallel -kP4 -n1 grep 1 ::: a b c d e f > out.par
  map -n1 -p 4 'grep 1' a b c d e f > out.map-unbuf
  map -n1 -p 4 'grep --line-buffered 1' a b c d e f > out.map-linebuf
  map -n1 -p 1 'grep --line-buffered 1' a b c d e f > out.map-serial
  ls -l out*
  md5sum out*

EXAMPLES FROM map's WEBSITE

Here are the examples from map's web page with the equivalent using GNU parallel:

  1$ ls *.gif | map convert % %B.png         # default max-args: 1

  1$ ls *.gif | parallel convert {} {.}.png

  2$ map "mkdir %B; tar -C %B -xf %" *.tgz   # default max-args: 1

  2$ parallel 'mkdir {.}; tar -C {.} -xf {}' :::  *.tgz

  3$ ls *.gif | map cp % /tmp                # default max-args: 100

  3$ ls *.gif | parallel -X cp {} /tmp

  4$ ls *.tar | map -n 1 tar -xf %

  4$ ls *.tar | parallel tar -xf

  5$ map "cp % /tmp" *.tgz

  5$ parallel cp {} /tmp ::: *.tgz

  6$ map "du -sm /home/%/mail" alice bob carol

  6$ parallel "du -sm /home/{}/mail" ::: alice bob carol
  or if you prefer running a single job with multiple args:
  6$ parallel -Xj1 "du -sm /home/{}/mail" ::: alice bob carol

  7$ cat /etc/passwd | map -d: 'echo user %1 has shell %7'

  7$ cat /etc/passwd | parallel --colsep : 'echo user {1} has shell {7}'

  8$ export MAP_MAX_PROCS=$(( `nproc` / 2 ))

  8$ export PARALLEL=-j50%

https://github.com/sitaramc/map (Last checked: 2020-05)

ladon can run multiple jobs on files in parallel.

ladon only works on files and the only way to specify files is using a quoted glob string (such as \*.jpg). It is not possible to list the files manually.

As replacement strings it uses FULLPATH DIRNAME BASENAME EXT RELDIR RELPATH

These can be simulated using GNU parallel by putting this in ~/.parallel/config:

  --rpl 'FULLPATH $_=Q($_);chomp($_=qx{readlink -f $_});'
  --rpl 'DIRNAME $_=Q(::dirname($_));chomp($_=qx{readlink -f $_});'
  --rpl 'BASENAME s:.*/::;s:\.[^/.]+$::;'
  --rpl 'EXT s:.*\.::'
  --rpl 'RELDIR $_=Q($_);chomp(($_,$c)=qx{readlink -f $_;pwd});
         s:\Q$c/\E::;$_=::dirname($_);'
  --rpl 'RELPATH $_=Q($_);chomp(($_,$c)=qx{readlink -f $_;pwd});
         s:\Q$c/\E::;'

ladon deals badly with filenames containing " and newline, and it fails for output larger than 200k:

  ladon '*' -- seq 36000 | wc

EXAMPLES FROM ladon MANUAL

It is assumed that the '--rpl's above are put in ~/.parallel/config and that it is run under a shell that supports '**' globbing (such as zsh):

  1$ ladon "**/*.txt" -- echo RELPATH

  1$ parallel echo RELPATH ::: **/*.txt

  2$ ladon "~/Documents/**/*.pdf" -- shasum FULLPATH >hashes.txt

  2$ parallel shasum FULLPATH ::: ~/Documents/**/*.pdf >hashes.txt

  3$ ladon -m thumbs/RELDIR "**/*.jpg" -- convert FULLPATH \
       -thumbnail 100x100^ -gravity center -extent 100x100 \
       thumbs/RELPATH

  3$ parallel mkdir -p thumbs/RELDIR\; convert FULLPATH
       -thumbnail 100x100^ -gravity center -extent 100x100 \
       thumbs/RELPATH ::: **/*.jpg

  4$ ladon "~/Music/*.wav" -- lame -V 2 FULLPATH DIRNAME/BASENAME.mp3

  4$ parallel lame -V 2 FULLPATH DIRNAME/BASENAME.mp3 ::: ~/Music/*.wav

https://github.com/danielgtaylor/ladon (Last checked: 2024-06)

Summary (see legend above):

I1 - - - - - I7

- - M3 - - (M6)

O1 O2 O3 - O5 O6 (O7) - - O10

E1 - - - - E6 -

- - - - - - - - -

- -

jobflow can run multiple jobs in parallel.

Just like xargs output from jobflow jobs running in parallel mix together by default. jobflow can buffer into files with -buffered (placed in /run/shm), but these are not cleaned up if jobflow dies unexpectedly (e.g. by Ctrl-C). If the total output is big (in the order of RAM+swap) it can cause the system to slow to a crawl and eventually run out of memory.

Just like xargs redirection and composed commands require wrapping with bash -c.

Input lines can at most be 4096 bytes.

jobflow is faster than GNU parallel but around 6 times slower than parallel-bash.

jobflow has no equivalent for --pipe, or --sshlogin.

jobflow makes it possible to set resource limits on the running jobs. This can be emulated by GNU parallel using bash's ulimit:

  jobflow -limits=mem=100M,cpu=3,fsize=20M,nofiles=300 myjob

  parallel 'ulimit -v 102400 -t 3 -f 204800 -n 300 myjob'

EXAMPLES FROM jobflow README

  1$ cat things.list | jobflow -threads=8 -exec ./mytask {}

  1$ cat things.list | parallel -j8 ./mytask {}

  2$ seq 100 | jobflow -threads=100 -exec echo {}

  2$ seq 100 | parallel -j100 echo {}

  3$ cat urls.txt | jobflow -threads=32 -exec wget {}

  3$ cat urls.txt | parallel -j32 wget {}

  4$ find . -name '*.bmp' | \
       jobflow -threads=8 -exec bmp2jpeg {.}.bmp {.}.jpg

  4$ find . -name '*.bmp' | \
       parallel -j8 bmp2jpeg {.}.bmp {.}.jpg

  5$ seq 100 | jobflow -skip 10 -count 10

  5$ seq 100 | parallel --filter '{1} > 10 and {1} <= 20' echo

  5$ seq 100 | parallel echo '{= $_>10 and $_<=20 or skip() =}'

https://github.com/rofl0r/jobflow (Last checked: 2024-06)

gargs can run multiple jobs in parallel.

Older versions cache output in memory. This causes it to be extremely slow when the output is larger than the physical RAM, and can cause the system to run out of memory.

See more details on this in man parallel_design.

Newer versions cache output in files, but leave files in $TMPDIR if it is killed.

Output to stderr (standard error) is changed if the command fails.

EXAMPLES FROM gargs WEBSITE

  1$ seq 12 -1 1 | gargs -p 4 -n 3 "sleep {0}; echo {1} {2}"

  1$ seq 12 -1 1 | parallel -P 4 -n 3 "sleep {1}; echo {2} {3}"

  2$ cat t.txt | gargs --sep "\s+" \
       -p 2 "echo '{0}:{1}-{2}' full-line: \'{}\'"

  2$ cat t.txt | parallel --colsep "\\s+" \
       -P 2 "echo '{1}:{2}-{3}' full-line: \'{}\'"

https://github.com/brentp/gargs (Last checked: 2016-08)

orgalorg can run the same job on multiple machines. This is related to --onall and --nonall.

orgalorg supports entering the SSH password - provided it is the same for all servers. GNU parallel advocates using ssh-agent instead, but it is possible to emulate orgalorg's behavior by setting SSHPASS and by using --ssh "sshpass ssh".

To make the emulation easier, make a simple alias:

  alias par_emul="parallel -j0 --ssh 'sshpass ssh' --nonall --tag --lb"

If you want to supply a password run:

  SSHPASS=`ssh-askpass`

or set the password directly:

  SSHPASS=P4$$w0rd!

If the above is set up you can then do:

  orgalorg -o frontend1 -o frontend2 -p -C uptime
  par_emul -S frontend1 -S frontend2 uptime

  orgalorg -o frontend1 -o frontend2 -p -C top -bid 1
  par_emul -S frontend1 -S frontend2 top -bid 1

  orgalorg -o frontend1 -o frontend2 -p -er /tmp -n \
    'md5sum /tmp/bigfile' -S bigfile
  par_emul -S frontend1 -S frontend2 --basefile bigfile \
    --workdir /tmp md5sum /tmp/bigfile

orgalorg has a progress indicator for the transferring of a file. GNU parallel does not.

https://github.com/reconquest/orgalorg (Last checked: 2016-08)

Rust parallel focuses on speed. It is almost as fast as xargs, but not as fast as parallel-bash. It implements a few features from GNU parallel, but lacks many functions. All these fail:

  # Read arguments from file
  parallel -a file echo
  # Changing the delimiter
  parallel -d _ echo ::: a_b_c_

These do something different from GNU parallel

  # -q to protect quoted $ and space
  parallel -q perl -e '$a=shift; print "$a"x100000000' ::: a b c
  # Generation of combination of inputs
  parallel echo {1} {2} ::: red green blue ::: S M L XL XXL
  # {= perl expression =} replacement string
  parallel echo '{= s/new/old/ =}' ::: my.new your.new
  # --pipe
  seq 100000 | parallel --pipe wc
  # linked arguments
  parallel echo ::: S M L :::+ sml med lrg ::: R G B :::+ red grn blu
  # Run different shell dialects
  zsh -c 'parallel echo \={} ::: zsh && true'
  csh -c 'parallel echo \$\{\} ::: shell && true'
  bash -c 'parallel echo \$\({}\) ::: pwd && true'
  # Rust parallel does not start before the last argument is read
  (seq 10; sleep 5; echo 2) | time parallel -j2 'sleep 2; echo'
  tail -f /var/log/syslog | parallel echo

Most of the examples from the book GNU Parallel 2018 do not work, thus Rust parallel is not close to being a compatible replacement.

Rust parallel has no remote facilities.

It uses /tmp/parallel for tmp files and does not clean up if terminated abruptly. If another user on the system uses Rust parallel, then /tmp/parallel will have the wrong permissions and Rust parallel will fail. A malicious user can setup the right permissions and symlink the output file to one of the user's files and next time the user uses Rust parallel it will overwrite this file.

  attacker$ mkdir /tmp/parallel
  attacker$ chmod a+rwX /tmp/parallel
  # Symlink to the file the attacker wants to zero out
  attacker$ ln -s ~victim/.important-file /tmp/parallel/stderr_1
  victim$ seq 1000 | parallel echo
  # This file is now overwritten with stderr from 'echo'
  victim$ cat ~victim/.important-file

If /tmp/parallel runs full during the run, Rust parallel does not report this, but finishes with success - thereby risking data loss.

https://github.com/mmstick/parallel (Last checked: 2016-08)

Summary (see legend above):

- (I2) - I4 - - -

M1 - M3 - - M6

- O2 O3 - O5 (O6) - x x

E1 - - (E4) E5 - - E8 ?

- - - - - - - - -

- -

I2: I was unable to cannot get parallelion to read from a file.

O6: There is extra output if a job fails.

E4: The default number of parallel jobs is the number of cpu threads.

-- is needed to force args not be parsed as options:

  parallelion 'echo {}' -- Runs without -v

  parallelion 'echo {}' Runs with -v

The commands are run through ion shell.

Ctrl-C does not stop processing.

The --log is similar to syslog - not a table.

The progressbar is nice.

parallelion is fast: 0.1 ms/job. Similar to parallel-bash.

EXAMPLES FROM parallelion

  1$ parallelion --progress 'echo {}' {1..1000}

  1$ parallel --bar echo {} ::: {1..1000}

  2$ parallelion --progress 'echo {}' $(seq 1 999)

  2$ seq 1 999 | parallel --bar echo

https://gitlab.redox-os.org/redox-os/parallel (Last checked: 2024-08)

rush (https://github.com/shenwei356/rush) is written in Go and based on gargs.

Just like GNU parallel rush buffers in temporary files. But opposite GNU parallel rush does not clean up, if the process dies abnormally.

rush has some string manipulations that can be emulated by putting this into ~/.parallel/config (/ is used instead of %, and % is used instead of ^ as that is closer to bash's ${var%postfix}):

  --rpl '{:} s:(\.[^/]+)*$::'
  --rpl '{:%([^}]+?)} s:$$1(\.[^/]+)*$::'
  --rpl '{/:%([^}]*?)} s:.*/(.*)$$1(\.[^/]+)*$:$1:'
  --rpl '{/:} s:(.*/)?([^/.]+)(\.[^/]+)*$:$2:'
  --rpl '{@(.*?)} /$$1/ and $_=$1;'

EXAMPLES FROM rush's WEBSITE

Here are the examples from rush's website with the equivalent command in GNU parallel.

1. Simple run, quoting is not necessary

  1$ seq 1 3 | rush echo {}

  1$ seq 1 3 | parallel echo {}

2. Read data from file (`-i`)

  2$ rush echo {} -i data1.txt -i data2.txt

  2$ cat data1.txt data2.txt | parallel echo {}

3. Keep output order (`-k`)

  3$ seq 1 3 | rush 'echo {}' -k

  3$ seq 1 3 | parallel -k echo {}

4. Timeout (`-t`)

  4$ time seq 1 | rush 'sleep 2; echo {}' -t 1

  4$ time seq 1 | parallel --timeout 1 'sleep 2; echo {}'

5. Retry (`-r`)

  5$ seq 1 | rush 'python unexisted_script.py' -r 1

  5$ seq 1 | parallel --retries 2 'python unexisted_script.py'

Use -u to see it is really run twice:

  5$ seq 1 | parallel -u --retries 2 'python unexisted_script.py'

6. Dirname (`{/}`) and basename (`{%}`) and remove custom suffix (`{^suffix}`)

  6$ echo dir/file_1.txt.gz | rush 'echo {/} {%} {^_1.txt.gz}'

  6$ echo dir/file_1.txt.gz |
       parallel --plus echo {//} {/} {%_1.txt.gz}

7. Get basename, and remove last (`{.}`) or any (`{:}`) extension

  7$ echo dir.d/file.txt.gz | rush 'echo {.} {:} {%.} {%:}'

  7$ echo dir.d/file.txt.gz | parallel 'echo {.} {:} {/.} {/:}'

8. Job ID, combine fields index and other replacement strings

  8$ echo 12 file.txt dir/s_1.fq.gz |
       rush 'echo job {#}: {2} {2.} {3%:^_1}'

  8$ echo 12 file.txt dir/s_1.fq.gz |
       parallel --colsep ' ' 'echo job {#}: {2} {2.} {3/:%_1}'

9. Capture submatch using regular expression (`{@regexp}`)

  9$ echo read_1.fq.gz | rush 'echo {@(.+)_\d}'

  9$ echo read_1.fq.gz | parallel 'echo {@(.+)_\d}'

10. Custom field delimiter (`-d`)

  10$ echo a=b=c | rush 'echo {1} {2} {3}' -d =

  10$ echo a=b=c | parallel -d = echo {1} {2} {3}

11. Send multi-lines to every command (`-n`)

  11$ seq 5 | rush -n 2 -k 'echo "{}"; echo'

  11$ seq 5 |
        parallel -n 2 -k \
          'echo {=-1 $_=join"\n",@arg[1..$#arg] =}; echo'

  11$ seq 5 | rush -n 2 -k 'echo "{}"; echo' -J ' '

  11$ seq 5 | parallel -n 2 -k 'echo {}; echo'

12. Custom record delimiter (`-D`), note that empty records are not used.

  12$ echo a b c d | rush -D " " -k 'echo {}'

  12$ echo a b c d | parallel -d " " -k 'echo {}'

  12$ echo abcd | rush -D "" -k 'echo {}'

  Cannot be done by GNU Parallel

  12$ cat fasta.fa
  >seq1
  tag
  >seq2
  cat
  gat
  >seq3
  attac
  a
  cat

  12$ cat fasta.fa | rush -D ">" \
        'echo FASTA record {#}: name: {1} sequence: {2}' -k -d "\n"
      # rush fails to join the multiline sequences

  12$ cat fasta.fa | (read -n1 ignore_first_char;
        parallel -d '>' --colsep '\n' echo FASTA record {#}: \
          name: {1} sequence: '{=2 $_=join"",@arg[2..$#arg]=}'
      )

13. Assign value to variable, like `awk -v` (`-v`)

  13$ seq 1 |
        rush 'echo Hello, {fname} {lname}!' -v fname=Wei -v lname=Shen

  13$ seq 1 |
        parallel -N0 \
          'fname=Wei; lname=Shen; echo Hello, ${fname} ${lname}!'

  13$ for var in a b; do \
  13$   seq 1 3 | rush -k -v var=$var 'echo var: {var}, data: {}'; \
  13$ done

In GNU parallel you would typically do:

  13$ seq 1 3 | parallel -k echo var: {1}, data: {2} ::: a b :::: -

If you really want the var:

  13$ seq 1 3 |
        parallel -k var={1} ';echo var: $var, data: {}' ::: a b :::: -

If you really want the for-loop:

  13$ for var in a b; do
        export var;
        seq 1 3 | parallel -k 'echo var: $var, data: {}';
      done

Contrary to rush this also works if the value is complex like:

  My brother's 12" records

14. Preset variable (`-v`), avoid repeatedly writing verbose replacement strings

  14$ # naive way
      echo read_1.fq.gz | rush 'echo {:^_1} {:^_1}_2.fq.gz'

  14$ echo read_1.fq.gz | parallel 'echo {:%_1} {:%_1}_2.fq.gz'

  14$ # macro + removing suffix
      echo read_1.fq.gz |
        rush -v p='{:^_1}' 'echo {p} {p}_2.fq.gz'

  14$ echo read_1.fq.gz |
        parallel 'p={:%_1}; echo $p ${p}_2.fq.gz'

  14$ # macro + regular expression
      echo read_1.fq.gz | rush -v p='{@(.+?)_\d}' 'echo {p} {p}_2.fq.gz'

  14$ echo read_1.fq.gz | parallel 'p={@(.+?)_\d}; echo $p ${p}_2.fq.gz'

Contrary to rush GNU parallel works with complex values:

  14$ echo "My brother's 12\"read_1.fq.gz" |
        parallel 'p={@(.+?)_\d}; echo $p ${p}_2.fq.gz'

15. Interrupt jobs by `Ctrl-C`, rush will stop unfinished commands and exit.

  15$ seq 1 20 | rush 'sleep 1; echo {}'
      ^C

  15$ seq 1 20 | parallel 'sleep 1; echo {}'
      ^C

16. Continue/resume jobs (`-c`). When some jobs failed (by execution failure, timeout, or canceling by user with `Ctrl + C`), please switch flag `-c/--continue` on and run again, so that `rush` can save successful commands and ignore them in NEXT run.

  16$ seq 1 3 | rush 'sleep {}; echo {}' -t 3 -c
      cat successful_cmds.rush
      seq 1 3 | rush 'sleep {}; echo {}' -t 3 -c

  16$ seq 1 3 | parallel --joblog mylog --timeout 2 \
        'sleep {}; echo {}'
      cat mylog
      seq 1 3 | parallel --joblog mylog --retry-failed \
        'sleep {}; echo {}'

Multi-line jobs:

  16$ seq 1 3 | rush 'sleep {}; echo {}; \
        echo finish {}' -t 3 -c -C finished.rush
      cat finished.rush
      seq 1 3 | rush 'sleep {}; echo {}; \
        echo finish {}' -t 3 -c -C finished.rush

  16$ seq 1 3 |
        parallel --joblog mylog --timeout 2 'sleep {}; echo {}; \
          echo finish {}'
      cat mylog
      seq 1 3 |
        parallel --joblog mylog --retry-failed 'sleep {}; echo {}; \
          echo finish {}'

17. A comprehensive example: downloading 1K+ pages given by three URL list files using `phantomjs save_page.js` (some page contents are dynamically generated by Javascript, so `wget` does not work). Here I set max jobs number (`-j`) as `20`, each job has a max running time (`-t`) of `60` seconds and `3` retry changes (`-r`). Continue flag `-c` is also switched on, so we can continue unfinished jobs. Luckily, it's accomplished in one run :)

  17$ for f in $(seq 2014 2016); do \
        /bin/rm -rf $f; mkdir -p $f; \
        cat $f.html.txt | rush -v d=$f -d = \
          'phantomjs save_page.js "{}" > {d}/{3}.html' \
          -j 20 -t 60 -r 3 -c; \
      done

GNU parallel can append to an existing joblog with '+':

  17$ rm mylog
      for f in $(seq 2014 2016); do
        /bin/rm -rf $f; mkdir -p $f;
        cat $f.html.txt |
          parallel -j20 --timeout 60 --retries 4 --joblog +mylog \
            --colsep = \
            phantomjs save_page.js {1}={2}={3} '>' $f/{3}.html
      done

18. A bioinformatics example: mapping with `bwa`, and processing result with `samtools`:

  18$ ref=ref/xxx.fa
      threads=25
      ls -d raw.cluster.clean.mapping/* \
        | rush -v ref=$ref -v j=$threads -v p='{}/{%}' \
        'bwa mem -t {j} -M -a {ref} {p}_1.fq.gz {p}_2.fq.gz >{p}.sam;\
        samtools view -bS {p}.sam > {p}.bam; \
        samtools sort -T {p}.tmp -@ {j} {p}.bam -o {p}.sorted.bam; \
        samtools index {p}.sorted.bam; \
        samtools flagstat {p}.sorted.bam > {p}.sorted.bam.flagstat; \
        /bin/rm {p}.bam {p}.sam;' \
        -j 2 --verbose -c -C mapping.rush

GNU parallel would use a function:

  18$ ref=ref/xxx.fa
      export ref
      thr=25
      export thr
      bwa_sam() {
        p="$1"
        bam="$p".bam
        sam="$p".sam
        sortbam="$p".sorted.bam
        bwa mem -t $thr -M -a $ref ${p}_1.fq.gz ${p}_2.fq.gz > "$sam"
        samtools view -bS "$sam" > "$bam"
        samtools sort -T ${p}.tmp -@ $thr "$bam" -o "$sortbam"
        samtools index "$sortbam"
        samtools flagstat "$sortbam" > "$sortbam".flagstat
        /bin/rm "$bam" "$sam"
      }
      export -f bwa_sam
      ls -d raw.cluster.clean.mapping/* |
        parallel -j 2 --verbose --joblog mylog bwa_sam

Other rush features

rush has:

• awk -v like custom defined variables (-v)

  With GNU parallel you would simply set a shell variable:

     parallel 'v={}; echo "$v"' ::: foo
     echo foo | rush -v v={} 'echo {v}'

  Also rush does not like special chars. So these do not work:

     echo does not work | rush -v v=\" 'echo {v}'
     echo "My  brother's  12\"  records" | rush -v v={} 'echo {v}'

  Whereas the corresponding GNU parallel version works:

     parallel 'v=\"; echo "$v"' ::: works
     parallel 'v={}; echo "$v"' ::: "My  brother's  12\"  records"

• Exit on first error(s) (-e)

  This is called --halt now,fail=1 (or shorter: --halt 2) when used with GNU parallel.

• Settable records sending to every command (-n, default 1)

  This is also called -n in GNU parallel.

• Practical replacement strings

  {:} remove any extension

  With GNU parallel this can be emulated by:

    parallel --plus echo '{/\..*/}' ::: foo.ext.bar.gz

  {^suffix}, remove suffix

  With GNU parallel this can be emulated by:

    parallel --plus echo '{%.bar.gz}' ::: foo.ext.bar.gz

  {@regexp}, capture submatch using regular expression

  With GNU parallel this can be emulated by:

    parallel --rpl '{@(.*?)} /$$1/ and $_=$1;' \
      echo '{@\d_(.*).gz}' ::: 1_foo.gz

  {%.}, {%:}, basename without extension

  With GNU parallel this can be emulated by:

    parallel echo '{= s:.*/::;s/\..*// =}' ::: dir/foo.bar.gz

  And if you need it often, you define a --rpl in $HOME/.parallel/config:

    --rpl '{%.} s:.*/::;s/\..*//'
    --rpl '{%:} s:.*/::;s/\..*//'

  Then you can use them as:

    parallel echo {%.} {%:} ::: dir/foo.bar.gz

• Preset variable (macro)

  E.g.

    echo foosuffix | rush -v p={^suffix} 'echo {p}_new_suffix'

  With GNU parallel this can be emulated by:

    echo foosuffix |
      parallel --plus 'p={%suffix}; echo ${p}_new_suffix'

  Opposite rush GNU parallel works fine if the input contains double space, ' and ":

    echo "1'6\"  foosuffix" |
      parallel --plus 'p={%suffix}; echo "${p}"_new_suffix'

• Commands of multi-lines

  While you can use multi-lined commands in GNU parallel, to improve readability GNU parallel discourages the use of multi-line commands. In most cases it can be written as a function:

    seq 1 3 |
      parallel --timeout 2 --joblog my.log 'sleep {}; echo {}; \
        echo finish {}'

  Could be written as:

    doit() {
      sleep "$1"
      echo "$1"
      echo finish "$1"
    }
    export -f doit
    seq 1 3 | parallel --timeout 2 --joblog my.log doit

  The failed commands can be resumed with:

    seq 1 3 |
      parallel --resume-failed --joblog my.log 'sleep {}; echo {};\
        echo finish {}'

https://github.com/shenwei356/rush (Last checked: 2017-05)

ClusterSSH solves a different problem than GNU parallel.

ClusterSSH opens a terminal window for each computer and using a master window you can run the same command on all the computers. This is typically used for administrating several computers that are almost identical.

GNU parallel runs the same (or different) commands with different arguments in parallel possibly using remote computers to help computing. If more than one computer is listed in -S GNU parallel may only use one of these (e.g. if there are 8 jobs to be run and one computer has 8 cores).

GNU parallel can be used as a poor-man's version of ClusterSSH:

  parallel --nonall -S server-a,server-b do_stuff foo bar

https://github.com/duncs/clusterssh (Last checked: 2024-06)

coshell only accepts full commands on standard input. Any quoting needs to be done by the user.

Commands are run in sh so any bash/tcsh/zsh specific syntax will not work.

Output can be buffered by using -d. Output is buffered in memory, so big output can cause swapping and therefore be terrible slow or even cause out of memory.

https://github.com/gdm85/coshell (Last checked: 2019-01)

Summary (see legend above):

- - - I4 - - I7

M1 - - - - -

O1 O2 O3 O4 O5 O6 - O8 - O10

- - - - - - -

- - - - - - - - -

- -

spread runs commands on all directories. It does not run jobs in parallel.

It can be emulated with GNU parallel using this Bash function:

  spread() {
    _cmds() {
      perl -e '$"=" && ";print "@ARGV"' "cd {}" "$@"
    }
    parallel $(_cmds "$@")'|| echo exit status $?' ::: */
  }

https://github.com/tfogo/spread (Last checked: 2024-06)

pyargs deals badly with input containing spaces. It buffers stdout, but not stderr. It buffers in RAM. {} does not work as replacement string. It does not support running functions.

pyargs does not support composed commands if run with --lines, and fails on pyargs traceroute gnu.org fsf.org.

Examples

  seq 5 | pyargs -P50 -L seq
  seq 5 | parallel -P50 --lb seq

  seq 5 | pyargs -P50 --mark -L seq
  seq 5 | parallel -P50 --lb \
    --tagstring OUTPUT'[{= $_=$job->replaced() =}]' seq
  # Similar, but not precisely the same
  seq 5 | parallel -P50 --lb --tag seq

  seq 5 | pyargs -P50  --mark command
  # Somewhat longer with GNU Parallel due to the special
  #   --mark formatting
  cmd="$(echo "command" | parallel --shellquote)"
  wrap_cmd() {
     echo "MARK $cmd $@================================" >&3
     echo "OUTPUT START[$cmd $@]:"
     eval $cmd "$@"
     echo "OUTPUT END[$cmd $@]"
  }
  (seq 5 | env_parallel -P2 wrap_cmd) 3>&1
  # Similar, but not exactly the same
  seq 5 | parallel -t --tag command

  (echo '1  2  3';echo 4 5 6) | pyargs  --stream seq
  (echo '1  2  3';echo 4 5 6) | perl -pe 's/\n/ /' |
    parallel -r -d' ' seq
  # Similar, but not exactly the same
  parallel seq ::: 1 2 3 4 5 6

https://github.com/robertblackwell/pyargs (Last checked: 2024-01)

concurrently runs jobs in parallel.

The output is prepended with the job number, and may be incomplete:

  $ concurrently 'seq 100000' | (sleep 3;wc -l)
  7165

When pretty printing it caches output in memory. Output mixes by using test MIX below whether or not output is cached.

There seems to be no way of making a template command and have concurrently fill that with different args. The full commands must be given on the command line.

There is also no way of controlling how many jobs should be run in parallel at a time - i.e. "number of jobslots". Instead all jobs are simply started in parallel.

https://github.com/kimmobrunfeldt/concurrently (Last checked: 2019-01)

map does not run jobs in parallel by default. The README suggests using:

  ... | map t 'sleep $t && say done &'

But this fails if more jobs are run in parallel than the number of available processes. Since there is no support for parallelization in map itself, the output also mixes:

  seq 10 | map i 'echo start-$i && sleep 0.$i && echo end-$i &'

The major difference is that GNU parallel is built for parallelization and map is not. So GNU parallel has lots of ways of dealing with the issues that parallelization raises:

• Keep the number of processes manageable
• Make sure output does not mix
• Make Ctrl-C kill all running processes

EXAMPLES FROM maps WEBSITE

Here are the 5 examples converted to GNU Parallel:

  1$ ls *.c | map f 'foo $f'
  1$ ls *.c | parallel foo

  2$ ls *.c | map f 'foo $f; bar $f'
  2$ ls *.c | parallel 'foo {}; bar {}'

  3$ cat urls | map u 'curl -O $u'
  3$ cat urls | parallel curl -O

  4$ printf "1\n1\n1\n" | map t 'sleep $t && say done'
  4$ printf "1\n1\n1\n" | parallel 'sleep {} && say done'
  4$ parallel 'sleep {} && say done' ::: 1 1 1

  5$ printf "1\n1\n1\n" | map t 'sleep $t && say done &'
  5$ printf "1\n1\n1\n" | parallel -j0 'sleep {} && say done'
  5$ parallel -j0 'sleep {} && say done' ::: 1 1 1

https://github.com/soveran/map (Last checked: 2019-01)

loop mixes stdout and stderr:

    loop 'ls /no-such-file' >/dev/null

loop's replacement string $ITEM does not quote strings:

    echo 'two  spaces' | loop 'echo $ITEM'

loop cannot run functions:

    myfunc() { echo joe; }
    export -f myfunc
    loop 'myfunc this fails'

EXAMPLES FROM loop's WEBSITE

Some of the examples from https://github.com/Miserlou/Loop/ can be emulated with GNU parallel:

    # A couple of functions will make the code easier to read
    $ loopy() {
        yes | parallel -uN0 -j1 "$@"
      }
    $ export -f loopy
    $ time_out() {
        parallel -uN0 -q --timeout "$@" ::: 1
      }
    $ match() {
        perl -0777 -ne 'grep /'"$1"'/,$_ and print or exit 1'
      }
    $ export -f match

    $ loop 'ls' --every 10s
    $ loopy --delay 10s ls

    $ loop 'touch $COUNT.txt' --count-by 5
    $ loopy touch '{= $_=seq()*5 =}'.txt

    $ loop --until-contains 200 -- \
        ./get_response_code.sh --site mysite.biz
    $ loopy --halt now,success=1 \
        './get_response_code.sh --site mysite.biz | match 200'

    $ loop './poke_server' --for-duration 8h
    $ time_out 8h loopy ./poke_server

    $ loop './poke_server' --until-success
    $ loopy --halt now,success=1 ./poke_server

    $ cat files_to_create.txt | loop 'touch $ITEM'
    $ cat files_to_create.txt | parallel touch {}

    $ loop 'ls' --for-duration 10min --summary
    # --joblog is somewhat more verbose than --summary
    $ time_out 10m loopy --joblog my.log ./poke_server; cat my.log

    $ loop 'echo hello'
    $ loopy echo hello

    $ loop 'echo $COUNT'
    # GNU Parallel counts from 1
    $ loopy echo {#}
    # Counting from 0 can be forced
    $ loopy echo '{= $_=seq()-1 =}'

    $ loop 'echo $COUNT' --count-by 2
    $ loopy echo '{= $_=2*(seq()-1) =}'

    $ loop 'echo $COUNT' --count-by 2 --offset 10
    $ loopy echo '{= $_=10+2*(seq()-1) =}'

    $ loop 'echo $COUNT' --count-by 1.1
    # GNU Parallel rounds 3.3000000000000003 to 3.3
    $ loopy echo '{= $_=1.1*(seq()-1) =}'

    $ loop 'echo $COUNT $ACTUALCOUNT' --count-by 2
    $ loopy echo '{= $_=2*(seq()-1) =} {#}'

    $ loop 'echo $COUNT' --num 3 --summary
    # --joblog is somewhat more verbose than --summary
    $ seq 3 | parallel --joblog my.log echo; cat my.log

    $ loop 'ls -foobarbatz' --num 3 --summary
    # --joblog is somewhat more verbose than --summary
    $ seq 3 | parallel --joblog my.log -N0 ls -foobarbatz; cat my.log

    $ loop 'echo $COUNT' --count-by 2 --num 50 --only-last
    # Can be emulated by running 2 jobs
    $ seq 49 | parallel echo '{= $_=2*(seq()-1) =}' >/dev/null
    $ echo 50| parallel echo '{= $_=2*(seq()-1) =}'

    $ loop 'date' --every 5s
    $ loopy --delay 5s date

    $ loop 'date' --for-duration 8s --every 2s
    $ time_out 8s loopy --delay 2s date

    $ loop 'date -u' --until-time '2018-05-25 20:50:00' --every 5s
    $ seconds=$((`date -d 2019-05-25T20:50:00 +%s` - `date  +%s`))s
    $ time_out $seconds loopy --delay 5s date -u

    $ loop 'echo $RANDOM' --until-contains "666"
    $ loopy --halt now,success=1 'echo $RANDOM | match 666'

    $ loop 'if (( RANDOM % 2 )); then
              (echo "TRUE"; true);
            else
              (echo "FALSE"; false);
            fi' --until-success
    $ loopy --halt now,success=1 'if (( $RANDOM % 2 )); then
                                    (echo "TRUE"; true);
                                  else
                                    (echo "FALSE"; false);
                                  fi'

    $ loop 'if (( RANDOM % 2 )); then
        (echo "TRUE"; true);
      else
        (echo "FALSE"; false);
      fi' --until-error
    $ loopy --halt now,fail=1 'if (( $RANDOM % 2 )); then
                                 (echo "TRUE"; true);
                               else
                                 (echo "FALSE"; false);
                               fi'

    $ loop 'date' --until-match "(\d{4})"
    $ loopy --halt now,success=1 'date | match [0-9][0-9][0-9][0-9]'

    $ loop 'echo $ITEM' --for red,green,blue
    $ parallel echo ::: red green blue

    $ cat /tmp/my-list-of-files-to-create.txt | loop 'touch $ITEM'
    $ cat /tmp/my-list-of-files-to-create.txt | parallel touch

    $ ls | loop 'cp $ITEM $ITEM.bak'; ls
    $ ls | parallel cp {} {}.bak; ls

    $ loop 'echo $ITEM | tr a-z A-Z' -i
    $ parallel 'echo {} | tr a-z A-Z'
    # Or more efficiently:
    $ parallel --pipe tr a-z A-Z

    $ loop 'echo $ITEM' --for "`ls`"
    $ parallel echo {} ::: "`ls`"

    $ ls | loop './my_program $ITEM' --until-success;
    $ ls | parallel --halt now,success=1 ./my_program {}

    $ ls | loop './my_program $ITEM' --until-fail;
    $ ls | parallel --halt now,fail=1 ./my_program {}

    $ ./deploy.sh;
      loop 'curl -sw "%{http_code}" http://coolwebsite.biz' \
        --every 5s --until-contains 200;
      ./announce_to_slack.sh
    $ ./deploy.sh;
      loopy --delay 5s --halt now,success=1 \
      'curl -sw "%{http_code}" http://coolwebsite.biz | match 200';
      ./announce_to_slack.sh

    $ loop "ping -c 1 mysite.com" --until-success; ./do_next_thing
    $ loopy --halt now,success=1 ping -c 1 mysite.com; ./do_next_thing

    $ ./create_big_file -o my_big_file.bin;
      loop 'ls' --until-contains 'my_big_file.bin';
      ./upload_big_file my_big_file.bin
    # inotifywait is a better tool to detect file system changes.
    # It can even make sure the file is complete
    # so you are not uploading an incomplete file
    $ inotifywait -qmre MOVED_TO -e CLOSE_WRITE --format %w%f . |
        grep my_big_file.bin

    $ ls | loop 'cp $ITEM $ITEM.bak'
    $ ls | parallel cp {} {}.bak

    $ loop './do_thing.sh' --every 15s --until-success --num 5
    $ parallel --retries 5 --delay 15s ::: ./do_thing.sh

https://github.com/Miserlou/Loop/ (Last checked: 2018-10)

lorikeet can run jobs in parallel. It does this based on a dependency graph described in a file, so this is similar to make.

https://github.com/cetra3/lorikeet (Last checked: 2018-10)

spp can run jobs in parallel. spp does not use a command template to generate the jobs, but requires jobs to be in a file. Output from the jobs mix.

https://github.com/john01dav/spp (Last checked: 2024-06)

paral prints a lot of status information and stores the output from the commands run into files. This means it cannot be used the middle of a pipe like this

  paral "echo this" "echo does not" "echo work" | wc

Instead it puts the output into files named like out_#_command.out.log. To get a very similar behaviour with GNU parallel use --results 'out_{#}_{=s/[^\sa-z_0-9]//g;s/\s+/_/g=}.log' --eta

paral only takes arguments on the command line and each argument should be a full command. Thus it does not use command templates.

This limits how many jobs it can run in total, because they all need to fit on a single command line.

paral has no support for running jobs remotely.

EXAMPLES FROM README.markdown

The examples from README.markdown and the corresponding command run with GNU parallel (--results 'out_{#}_{=s/[^\sa-z_0-9]//g;s/\s+/_/g=}.log' --eta is omitted from the GNU parallel command):

  1$ paral "command 1" "command 2 --flag" "command arg1 arg2"
  1$ parallel ::: "command 1" "command 2 --flag" "command arg1 arg2"

  2$ paral "sleep 1 && echo c1" "sleep 2 && echo c2" \
       "sleep 3 && echo c3" "sleep 4 && echo c4"  "sleep 5 && echo c5"
  2$ parallel ::: "sleep 1 && echo c1" "sleep 2 && echo c2" \
       "sleep 3 && echo c3" "sleep 4 && echo c4"  "sleep 5 && echo c5"
     # Or shorter:
     parallel "sleep {} && echo c{}" ::: {1..5}

  3$ paral -n=0 "sleep 5 && echo c5" "sleep 4 && echo c4" \
       "sleep 3 && echo c3" "sleep 2 && echo c2" "sleep 1 && echo c1"
  3$ parallel ::: "sleep 5 && echo c5" "sleep 4 && echo c4" \
       "sleep 3 && echo c3" "sleep 2 && echo c2" "sleep 1 && echo c1"
     # Or shorter:
     parallel -j0 "sleep {} && echo c{}" ::: 5 4 3 2 1

  4$ paral -n=1 "sleep 5 && echo c5" "sleep 4 && echo c4" \
       "sleep 3 && echo c3" "sleep 2 && echo c2" "sleep 1 && echo c1"
  4$ parallel -j1 "sleep {} && echo c{}" ::: 5 4 3 2 1

  5$ paral -n=2 "sleep 5 && echo c5" "sleep 4 && echo c4" \
       "sleep 3 && echo c3" "sleep 2 && echo c2" "sleep 1 && echo c1"
  5$ parallel -j2 "sleep {} && echo c{}" ::: 5 4 3 2 1

  6$ paral -n=5 "sleep 5 && echo c5" "sleep 4 && echo c4" \
       "sleep 3 && echo c3" "sleep 2 && echo c2" "sleep 1 && echo c1"
  6$ parallel -j5 "sleep {} && echo c{}" ::: 5 4 3 2 1

  7$ paral -n=1 "echo a && sleep 0.5 && echo b && sleep 0.5 && \
       echo c && sleep 0.5 && echo d && sleep 0.5 && \
       echo e && sleep 0.5 && echo f && sleep 0.5 && \
       echo g && sleep 0.5 && echo h"
  7$ parallel ::: "echo a && sleep 0.5 && echo b && sleep 0.5 && \
       echo c && sleep 0.5 && echo d && sleep 0.5 && \
       echo e && sleep 0.5 && echo f && sleep 0.5 && \
       echo g && sleep 0.5 && echo h"

https://github.com/amattn/paral (Last checked: 2024-06)

concurr is built to run jobs in parallel using a client/server model.

EXAMPLES FROM README.md

The examples from README.md:

  1$ concurr 'echo job {#} on slot {%}: {}' : arg1 arg2 arg3 arg4
  1$ parallel 'echo job {#} on slot {%}: {}' ::: arg1 arg2 arg3 arg4

  2$ concurr 'echo job {#} on slot {%}: {}' :: file1 file2 file3
  2$ parallel 'echo job {#} on slot {%}: {}' :::: file1 file2 file3

  3$ concurr 'echo {}' < input_file
  3$ parallel 'echo {}' < input_file

  4$ cat file | concurr 'echo {}'
  4$ cat file | parallel 'echo {}'

concurr deals badly empty input files and with output larger than 64 KB.

https://github.com/mmstick/concurr (Last checked: 2024-01)

lesser-parallel is the inspiration for parallel --embed. Both lesser-parallel and parallel --embed define bash functions that can be included as part of a bash script to run jobs in parallel.

lesser-parallel implements a few of the replacement strings, but hardly any options, whereas parallel --embed gives you the full GNU parallel experience.

https://github.com/kou1okada/lesser-parallel (Last checked: 2024-06)

npm-parallel can run npm tasks in parallel.

There are no examples and very little documentation, so it is hard to compare to GNU parallel.

https://github.com/spion/npm-parallel (Last checked: 2024-06)

machma runs tasks in parallel. It gives time stamped output. It buffers in RAM.

EXAMPLES FROM README.md

The examples from README.md:

  1$ # Put shorthand for timestamp in config for the examples
     echo '--rpl '\
       \''{time} $_=::strftime("%Y-%m-%d %H:%M:%S",localtime())'\' \
       > ~/.parallel/machma
     echo '--line-buffer --tagstring "{#} {time} {}"' \
       >> ~/.parallel/machma

  2$ find . -iname '*.jpg' |
       machma --  mogrify -resize 1200x1200 -filter Lanczos {}
     find . -iname '*.jpg' |
       parallel --bar -Jmachma mogrify -resize 1200x1200 \
         -filter Lanczos {}

  3$ cat /tmp/ips | machma -p 2 -- ping -c 2 -q {}
  3$ cat /tmp/ips | parallel -j2 -Jmachma ping -c 2 -q {}

  4$ cat /tmp/ips |
       machma -- sh -c 'ping -c 2 -q $0 > /dev/null && echo alive' {}
  4$ cat /tmp/ips |
       parallel -Jmachma 'ping -c 2 -q {} > /dev/null && echo alive'

  5$ find . -iname '*.jpg' |
       machma --timeout 5s -- mogrify -resize 1200x1200 \
         -filter Lanczos {}
  5$ find . -iname '*.jpg' |
       parallel --timeout 5s --bar mogrify -resize 1200x1200 \
         -filter Lanczos {}

  6$ find . -iname '*.jpg' -print0 |
       machma --null --  mogrify -resize 1200x1200 -filter Lanczos {}
  6$ find . -iname '*.jpg' -print0 |
       parallel --null --bar mogrify -resize 1200x1200 \
         -filter Lanczos {}

https://github.com/fd0/machma (Last checked: 2019-06)

Summary (see legend above):

- I2 I3 I4 - - -

M1 - M3 - - M6

- O2 O3 - - - - x x

E1 E2 - - - - -

- - - - - - - - -

- -

interlace is built for network analysis to run network tools in parallel.

interface does not buffer output, so output from different jobs mixes.

The overhead for each target is O(n*n), so with 1000 targets it becomes very slow with an overhead in the order of 500ms/target.

EXAMPLES FROM interlace's WEBSITE

Using prips most of the examples from https://github.com/codingo/Interlace can be run with GNU parallel:

Blocker

  commands.txt:
    mkdir -p _output_/_target_/scans/
    _blocker_
    nmap _target_ -oA _output_/_target_/scans/_target_-nmap
  interlace -tL ./targets.txt -cL commands.txt -o $output

  parallel -a targets.txt \
    mkdir -p $output/{}/scans/\; nmap {} -oA $output/{}/scans/{}-nmap

Blocks

  commands.txt:
    _block:nmap_
    mkdir -p _target_/output/scans/
    nmap _target_ -oN _target_/output/scans/_target_-nmap
    _block:nmap_
    nikto --host _target_
  interlace -tL ./targets.txt -cL commands.txt

  _nmap() {
    mkdir -p $1/output/scans/
    nmap $1 -oN $1/output/scans/$1-nmap
  }
  export -f _nmap
  parallel ::: _nmap "nikto --host" :::: targets.txt

Run Nikto Over Multiple Sites

  interlace -tL ./targets.txt -threads 5 \
    -c "nikto --host _target_ > ./_target_-nikto.txt" -v

  parallel -a targets.txt -P5 nikto --host {} \> ./{}_-nikto.txt

Run Nikto Over Multiple Sites and Ports

  interlace -tL ./targets.txt -threads 5 -c \
    "nikto --host _target_:_port_ > ./_target_-_port_-nikto.txt" \
    -p 80,443 -v

  parallel -P5 nikto --host {1}:{2} \> ./{1}-{2}-nikto.txt \
    :::: targets.txt ::: 80 443

Run a List of Commands against Target Hosts

  commands.txt:
    nikto --host _target_:_port_ > _output_/_target_-nikto.txt
    sslscan _target_:_port_ >  _output_/_target_-sslscan.txt
    testssl.sh _target_:_port_ > _output_/_target_-testssl.txt
  interlace -t example.com -o ~/Engagements/example/ \
    -cL ./commands.txt -p 80,443

  parallel --results ~/Engagements/example/{2}:{3}{1} {1} {2}:{3} \
    ::: "nikto --host" sslscan testssl.sh ::: example.com ::: 80 443

CIDR notation with an application that doesn't support it

  interlace -t 192.168.12.0/24 -c "vhostscan _target_ \
    -oN _output_/_target_-vhosts.txt" -o ~/scans/ -threads 50

  prips 192.168.12.0/24 |
    parallel -P50 vhostscan {} -oN ~/scans/{}-vhosts.txt

Glob notation with an application that doesn't support it

  interlace -t 192.168.12.* -c "vhostscan _target_ \
    -oN _output_/_target_-vhosts.txt" -o ~/scans/ -threads 50

  # Glob is not supported in prips
  prips 192.168.12.0/24 |
    parallel -P50 vhostscan {} -oN ~/scans/{}-vhosts.txt

Dash (-) notation with an application that doesn't support it

  interlace -t 192.168.12.1-15 -c \
    "vhostscan _target_ -oN _output_/_target_-vhosts.txt" \
    -o ~/scans/ -threads 50

  # Dash notation is not supported in prips
  prips 192.168.12.1 192.168.12.15 |
    parallel -P50 vhostscan {} -oN ~/scans/{}-vhosts.txt

Threading Support for an application that doesn't support it

  interlace -tL ./target-list.txt -c \
    "vhostscan -t _target_ -oN _output_/_target_-vhosts.txt" \
    -o ~/scans/ -threads 50

  cat ./target-list.txt |
    parallel -P50 vhostscan -t {} -oN ~/scans/{}-vhosts.txt

alternatively

  ./vhosts-commands.txt:
    vhostscan -t $target -oN _output_/_target_-vhosts.txt
  interlace -cL ./vhosts-commands.txt -tL ./target-list.txt \
    -threads 50 -o ~/scans

  ./vhosts-commands.txt:
    vhostscan -t "$1" -oN "$2"
  parallel -P50 ./vhosts-commands.txt {} ~/scans/{}-vhosts.txt \
    :::: ./target-list.txt

Exclusions

  interlace -t 192.168.12.0/24 -e 192.168.12.0/26 -c \
    "vhostscan _target_ -oN _output_/_target_-vhosts.txt" \
    -o ~/scans/ -threads 50

  prips 192.168.12.0/24 | grep -xv -Ff <(prips 192.168.12.0/26) |
    parallel -P50 vhostscan {} -oN ~/scans/{}-vhosts.txt

Run Nikto Using Multiple Proxies

   interlace -tL ./targets.txt -pL ./proxies.txt -threads 5 -c \
     "nikto --host _target_:_port_ -useproxy _proxy_ > \
      ./_target_-_port_-nikto.txt" -p 80,443 -v

   parallel -j5 \
     "nikto --host {1}:{2} -useproxy {3} > ./{1}-{2}-nikto.txt" \
     :::: ./targets.txt ::: 80 443 :::: ./proxies.txt

https://github.com/codingo/Interlace (Last checked: 2019-09)

I have been unable to get the code to run at all. It seems unfinished.

https://github.com/otonvm/Parallel (Last checked: 2024-06)

par requires Haskell to work. This limits the number of platforms this can work on.

par does line buffering in memory. The memory usage is 3x the longest line (compared to 1x for parallel --lb). Commands must be given as arguments. There is no template.

These are the examples from https://github.com/k-bx/par with the corresponding GNU parallel command.

  par "echo foo; sleep 1; echo foo; sleep 1; echo foo" \
      "echo bar; sleep 1; echo bar; sleep 1; echo bar" && echo "success"
  parallel --lb ::: "echo foo; sleep 1; echo foo; sleep 1; echo foo" \
      "echo bar; sleep 1; echo bar; sleep 1; echo bar" && echo "success"

  par "echo foo; sleep 1; foofoo" \
      "echo bar; sleep 1; echo bar; sleep 1; echo bar" && echo "success"
  parallel --lb --halt 1 ::: "echo foo; sleep 1; foofoo" \
      "echo bar; sleep 1; echo bar; sleep 1; echo bar" && echo "success"

  par "PARPREFIX=[fooechoer] echo foo" "PARPREFIX=[bar] echo bar"
  parallel --lb --colsep , --tagstring {1} {2} \
    ::: "[fooechoer],echo foo" "[bar],echo bar"

  par --succeed "foo" "bar" && echo 'wow'
  parallel "foo" "bar"; true && echo 'wow'

https://github.com/k-bx/par (Last checked: 2019-02)

parallelshell does not allow for composed commands:

  # This does not work
  parallelshell 'echo foo;echo bar' 'echo baz;echo quuz'

Instead you have to wrap that in a shell:

  parallelshell 'sh -c "echo foo;echo bar"' 'sh -c "echo baz;echo quuz"'

It buffers output in RAM. All commands must be given on the command line and all commands are started in parallel at the same time. This will cause the system to freeze if there are so many jobs that there is not enough memory to run them all at the same time.

https://github.com/keithamus/parallelshell (Last checked: 2019-02)

https://github.com/darkguy2008/parallelshell (Last checked: 2019-03)

shell-executor does not allow for composed commands:

  # This does not work
  sx 'echo foo;echo bar' 'echo baz;echo quuz'

Instead you have to wrap that in a shell:

  sx 'sh -c "echo foo;echo bar"' 'sh -c "echo baz;echo quuz"'

It buffers output in RAM. All commands must be given on the command line and all commands are started in parallel at the same time. This will cause the system to freeze if there are so many jobs that there is not enough memory to run them all at the same time.

https://github.com/royriojas/shell-executor (Last checked: 2024-06)

par buffers in memory to avoid mixing of jobs. It takes 1s per 1 million output lines.

par needs to have all commands before starting the first job. The jobs are read from stdin (standard input) so any quoting will have to be done by the user.

Stdout (standard output) is prepended with o:. Stderr (standard error) is sendt to stdout (standard output) and prepended with e:.

For short jobs with little output par is 20% faster than GNU parallel and 60% slower than xargs.

https://github.com/UnixJunkie/PAR

https://savannah.nongnu.org/projects/par (Last checked: 2019-02)

fd does not support composed commands, so commands must be wrapped in sh -c.

It buffers output in RAM.

It only takes file names from the filesystem as input (similar to find).

https://github.com/sharkdp/fd (Last checked: 2019-02)

lateral is very similar to sem: It takes a single command and runs it in the background. The design means that output from parallel running jobs may mix. If it dies unexpectly it leaves a socket in ~/.lateral/socket.PID.

lateral deals badly with too long command lines. This makes the lateral server crash:

  lateral run echo `seq 100000| head -c 1000k`

Any options will be read by lateral so this does not work (lateral interprets the -l):

  lateral run ls -l

Composed commands do not work:

  lateral run pwd ';' ls

Functions do not work:

  myfunc() { echo a; }
  export -f myfunc
  lateral run myfunc

Running emacs in the terminal causes the parent shell to die:

  echo '#!/bin/bash' > mycmd
  echo emacs -nw >> mycmd
  chmod +x mycmd
  lateral start
  lateral run ./mycmd

Here are the examples from https://github.com/akramer/lateral with the corresponding GNU sem and GNU parallel commands:

  1$ lateral start
     for i in $(cat /tmp/names); do
       lateral run -- some_command $i
     done
     lateral wait
  
  1$ for i in $(cat /tmp/names); do
       sem some_command $i
     done
     sem --wait
  
  1$ parallel some_command :::: /tmp/names

  2$ lateral start
     for i in $(seq 1 100); do
       lateral run -- my_slow_command < workfile$i > /tmp/logfile$i
     done
     lateral wait
    
  2$ for i in $(seq 1 100); do
       sem my_slow_command < workfile$i > /tmp/logfile$i
     done
     sem --wait
    
  2$ parallel 'my_slow_command < workfile{} > /tmp/logfile{}' \
       ::: {1..100}

  3$ lateral start -p 0 # yup, it will just queue tasks
     for i in $(seq 1 100); do
       lateral run -- command_still_outputs_but_wont_spam inputfile$i
     done
     # command output spam can commence
     lateral config -p 10; lateral wait
    
  3$ for i in $(seq 1 100); do
       echo "command inputfile$i" >> joblist
     done
     parallel -j 10 :::: joblist
  
  3$ echo 1 > /tmp/njobs
     parallel -j /tmp/njobs command inputfile{} \
       ::: {1..100} &
     echo 10 >/tmp/njobs
     wait

https://github.com/akramer/lateral (Last checked: 2024-06)

The examples from https://github.com/amritb/with-this.git and the corresponding GNU parallel command:

  with -v "$(cat myurls.txt)" "curl -L this"
  parallel curl -L ::: myurls.txt

  with -v "$(cat myregions.txt)" \
    "aws --region=this ec2 describe-instance-status"
  parallel aws --region={} ec2 describe-instance-status \
    :::: myregions.txt

  with -v "$(ls)" "kubectl --kubeconfig=this get pods"
  ls | parallel kubectl --kubeconfig={} get pods

  with -v "$(ls | grep config)" "kubectl --kubeconfig=this get pods"
  ls | grep config | parallel kubectl --kubeconfig={} get pods

  with -v "$(echo {1..10})" "echo 123"
  parallel -N0 echo 123 ::: {1..10}

Stderr is merged with stdout. with-this buffers in RAM. It uses 3x the output size, so you cannot have output larger than 1/3rd the amount of RAM. The input values cannot contain spaces. Composed commands do not work.

with-this gives some additional information, so the output has to be cleaned before piping it to the next command.

https://github.com/amritb/with-this.git (Last checked: 2024-06)

Summary (see legend above):

- - - I4 - - I7

- - M3 - - M6

- O2 O3 - O5 O6 - x x

E1 - - - - - E7

- x x x x x x x x

- -

EXAMPLES FROM Tollef's parallel MANUAL

Tollef parallel sh -c "echo hi; sleep 2; echo bye" -- 1 2 3

GNU parallel "echo hi; sleep 2; echo bye" ::: 1 2 3

Tollef parallel -j 3 ufraw -o processed -- *.NEF

GNU parallel -j 3 ufraw -o processed ::: *.NEF

Tollef parallel -j 3 -- ls df "echo hi"

GNU parallel -j 3 ::: ls df "echo hi"

(Last checked: 2019-08)

Summary (see legend above):

I1 - - - - - I7

- - M3 M4 - -

- O2 O3 - O5 O6 - O8 -

E1 - - E4 - - -

- - - - - - - - -

- -

rargs has elegant ways of doing named regexp capture and field ranges.

With GNU parallel you can use --rpl to get a similar functionality as regexp capture gives, and use join and @arg to get the field ranges. But the syntax is longer. This:

  --rpl '{r(\d+)\.\.(\d+)} $_=join"$opt::colsep",@arg[$$1..$$2]'

would make it possible to use:

  {1r3..6}

for field 3..6.

For full support of {n..m:s} including negative numbers use a dynamic replacement string like this:

  PARALLEL=--rpl\ \''{r((-?\d+)?)\.\.((-?\d+)?)((:([^}]*))?)}
          $a = defined $$2 ? $$2 < 0 ? 1+$#arg+$$2 : $$2 : 1;
          $b = defined $$4 ? $$4 < 0 ? 1+$#arg+$$4 : $$4 : $#arg+1;
          $s = defined $$6 ? $$7 : " ";
          $_ = join $s,@arg[$a..$b]'\'
  export PARALLEL

You can then do:

  head /etc/passwd | parallel --colsep : echo ..={1r..} ..3={1r..3} \
    4..={1r4..} 2..4={1r2..4} 3..3={1r3..3} ..3:-={1r..3:-} \
    ..3:/={1r..3:/} -1={-1} -5={-5} -6={-6} -3..={1r-3..}

EXAMPLES FROM rargs MANUAL

  1$ ls *.bak | rargs -p '(.*)\.bak' mv {0} {1}

  1$ ls *.bak | parallel mv {} {.}

  2$ cat download-list.csv |
       rargs -p '(?P<url>.*),(?P<filename>.*)' wget {url} -O {filename}

  2$ cat download-list.csv |
       parallel --csv wget {1} -O {2}
  # or use regexps:
  2$ cat download-list.csv |
       parallel --rpl '{url} s/,.*//' --rpl '{filename} s/.*?,//' \
         wget {url} -O {filename}

  3$ cat /etc/passwd |
       rargs -d: echo -e 'id: "{1}"\t name: "{5}"\t rest: "{6..::}"'

  3$ cat /etc/passwd |
       parallel -q --colsep : \
         echo -e 'id: "{1}"\t name: "{5}"\t rest: "{=6 $_=join":",@arg[6..$#arg]=}"'

https://github.com/lotabout/rargs (Last checked: 2020-01)

Summary (see legend above):

I1 - - - - - -

M1 - M3 - - M6

O1 - O3 - O5 - - x x

E1 - - E4 - - -

- - - - - - - - -

- -

Newline separates arguments, but newline at the end of file is treated as an empty argument. So this runs 2 jobs:

  echo two_jobs | threader -run 'echo "$THREADID"'

threader ignores stderr, so any output to stderr is lost. threader buffers in RAM, so output bigger than the machine's virtual memory will cause the machine to crash.

https://github.com/voodooEntity/threader (Last checked: 2024-06)

Summary (see legend above):

I1 I2 - - - - -

M1 - (M3) - - M6

O1 O2 O3 - O5 O6 - x x -

E1 - - - - - -

- - - - - - - - -

- -

(M3): You can add a prefix and a postfix to the input, so it means you can only insert the argument on the command line once.

runp runs 10 jobs in parallel by default.  runp blocks if output of a command is > 64 Kbytes.  Quoting of input is needed.  It adds output to stderr (this can be prevented with -q)

Examples as GNU Parallel

  base='https://images-api.nasa.gov/search'
  query='jupiter'
  desc='planet'
  type='image'
  url="$base?q=$query&description=$desc&media_type=$type"
  
  # Download the images in parallel using runp
  curl -s $url | jq -r .collection.items[].href | \
    runp -p 'curl -s' | jq -r .[] | grep large | \
    runp -p 'curl -s -L -O'

  time curl -s $url | jq -r .collection.items[].href | \
    runp -g 1 -q -p 'curl -s' | jq -r .[] | grep large | \
    runp -g 1 -q -p 'curl -s -L -O'

  # Download the images in parallel
  curl -s $url | jq -r .collection.items[].href | \
    parallel curl -s | jq -r .[] | grep large | \
    parallel curl -s -L -O
  
  time curl -s $url | jq -r .collection.items[].href | \
    parallel -j 1 curl -s | jq -r .[] | grep large | \
    parallel -j 1 curl -s -L -O

Run some test commands (read from file)

  # Create a file containing commands to run in parallel.
  cat << EOF > /tmp/test-commands.txt
  sleep 5
  sleep 3
  blah     # this will fail
  ls $PWD  # PWD shell variable is used here
  EOF
  
  # Run commands from the file.
  runp /tmp/test-commands.txt > /dev/null
  
  parallel -a /tmp/test-commands.txt > /dev/null

Ping several hosts and see packet loss (read from stdin)

  # First copy this line and press Enter
  runp -p 'ping -c 5 -W 2' -s '| grep loss'
  localhost
  1.1.1.1
  8.8.8.8
  # Press Enter and Ctrl-D when done entering the hosts

  # First copy this line and press Enter
  parallel ping -c 5 -W 2 {} '| grep loss'
  localhost
  1.1.1.1
  8.8.8.8
  # Press Enter and Ctrl-D when done entering the hosts

Get directories' sizes (read from stdin)

  echo -e "$HOME\n/etc\n/tmp" | runp -q -p 'sudo du -sh'

  echo -e "$HOME\n/etc\n/tmp" | parallel sudo du -sh
  # or:
  parallel sudo du -sh ::: "$HOME" /etc /tmp

Compress files

  find . -iname '*.txt' | runp -p 'gzip --best'

  find . -iname '*.txt' | parallel gzip --best

Measure HTTP request + response time

  export CURL="curl -w 'time_total:  %{time_total}\n'"
  CURL="$CURL -o /dev/null -s https://golang.org/"
  perl -wE 'for (1..10) { say $ENV{CURL} }' |
     runp -q  # Make 10 requests

  perl -wE 'for (1..10) { say $ENV{CURL} }' | parallel
  # or:
  parallel -N0 "$CURL" ::: {1..10}

Find open TCP ports

  cat << EOF > /tmp/host-port.txt
  localhost 22
  localhost 80
  localhost 81
  127.0.0.1 443
  127.0.0.1 444
  scanme.nmap.org 22
  scanme.nmap.org 23
  scanme.nmap.org 443
  EOF
  
  1$ cat /tmp/host-port.txt |
       runp -q -p 'netcat -v -w2 -z' 2>&1 | egrep '(succeeded!|open)$'
  
  # --colsep is needed to split the line
  1$ cat /tmp/host-port.txt |
       parallel --colsep ' ' netcat -v -w2 -z 2>&1 |
       egrep '(succeeded!|open)$'
  # or use uq for unquoted:
  1$ cat /tmp/host-port.txt |
       parallel netcat -v -w2 -z {=uq=} 2>&1 |
       egrep '(succeeded!|open)$'

https://github.com/jreisinger/runp (Last checked: 2020-04)

Summary (see legend above):

- - - I4 - - -

M1 - M3 - - M6

- - O3 - O5 - - x x O10

E1 - - E4 - - -

- - - - - - - - -

- -

papply does not print the output if the command fails:

  $ papply 'echo %F; false' foo
  "echo foo; false" did not succeed

papply's replacement strings (%F %d %f %n %e %z) can be simulated in GNU parallel by putting this in ~/.parallel/config:

  --rpl '%F'
  --rpl '%d $_=Q(::dirname($_));'
  --rpl '%f s:.*/::;'
  --rpl '%n s:.*/::;s:\.[^/.]+$::;'
  --rpl '%e s:.*\.:.:'
  --rpl '%z $_=""'

papply buffers in RAM, and uses twice the amount of output. So output of 5 GB takes 10 GB RAM.

The buffering is very CPU intensive: Buffering a line of 5 GB takes 40 seconds (compared to 10 seconds with GNU parallel).

Examples as GNU Parallel

  1$ papply gzip *.txt
  
  1$ parallel gzip ::: *.txt
  
  2$ papply "convert %F %n.jpg" *.png
  
  2$ parallel convert {} {.}.jpg ::: *.png

https://pypi.org/project/papply/ (Last checked: 2020-04)

Summary (see legend above):

- - - I4 - - I7

- - - - - M6

- O2 O3 - O5 O6 - x x O10

E1 - - E4 - E6 -

- - - - - - - - -

S1 S2

async is very similary to GNU parallel's --semaphore mode (aka sem). async requires the user to start a server process.

The input is quoted like -q so you need bash -c "...;..." to run composed commands.

Examples as GNU Parallel

  1$ S="/tmp/example_socket"
  
  1$ ID=myid
  
  2$ async -s="$S" server --start
  
  2$ # GNU Parallel does not need a server to run
  
  3$ for i in {1..20}; do
         # prints command output to stdout
         async -s="$S" cmd -- bash -c "sleep 1 && echo test $i"
     done
  
  3$ for i in {1..20}; do
         # prints command output to stdout
         sem --id "$ID" -j100% "sleep 1 && echo test $i"
         # GNU Parallel will only print job when it is done
         # If you need output from different jobs to mix
         # use -u or --line-buffer
         sem --id "$ID" -j100% --line-buffer "sleep 1 && echo test $i"
     done
  
  4$ # wait until all commands are finished
     async -s="$S" wait
  
  4$ sem --id "$ID" --wait
  
  5$ # configure the server to run four commands in parallel
     async -s="$S" server -j4
  
  5$ export PARALLEL=-j4
  
  6$ mkdir "/tmp/ex_dir"
     for i in {21..40}; do
       # redirects command output to /tmp/ex_dir/file*
       async -s="$S" cmd -o "/tmp/ex_dir/file$i" -- \
         bash -c "sleep 1 && echo test $i"
     done
  
  6$ mkdir "/tmp/ex_dir"
     for i in {21..40}; do
       # redirects command output to /tmp/ex_dir/file*
       sem --id "$ID" --result '/tmp/my-ex/file-{=$_=""=}'"$i" \
         "sleep 1 && echo test $i"
     done
  
  7$ sem --id "$ID" --wait
  
  7$ async -s="$S" wait
  
  8$ # stops server
     async -s="$S" server --stop
  
  8$ # GNU Parallel does not need to stop a server

https://github.com/ctbur/async/ (Last checked: 2024-06)

Summary (see legend above):

I1 I2 - - - - I7

M1 - - - - M6

O1 O2 O3 O4 O5 - O7 - - O10

E1 - - E4 - - -

- - - - - - - - -

- -

pardi is very similar to parallel --pipe --cat: It reads blocks of data and not arguments. So it cannot insert an argument in the command line. It puts the block into a temporary file, and this file name (%IN) can be put in the command line. You can only use %IN once.

It can also run full command lines in parallel (like: cat file | parallel).

EXAMPLES FROM pardi test.sh

  1$ time pardi -v -c 100 -i data/decoys.smi -ie .smi -oe .smi \
       -o data/decoys_std_pardi.smi \
          -w '(standardiser -i %IN -o %OUT 2>&1) > /dev/null'
  
  1$ cat data/decoys.smi |
       time parallel -N 100 --pipe --cat \
         '(standardiser -i {} -o {#} 2>&1) > /dev/null; cat {#}; rm {#}' \
         > data/decoys_std_pardi.smi
  
  2$ pardi -n 1 -i data/test_in.types -o data/test_out.types \
             -d 'r:^#atoms:' -w 'cat %IN > %OUT'
  
  2$ cat data/test_in.types |
       parallel -n 1 -k --pipe --cat --regexp --recstart '^#atoms' \
         'cat {}' > data/test_out.types
  
  3$ pardi -c 6 -i data/test_in.types -o data/test_out.types \
             -d 'r:^#atoms:' -w 'cat %IN > %OUT'
  
  3$ cat data/test_in.types |
       parallel -n 6 -k --pipe --cat --regexp --recstart '^#atoms' \
         'cat {}' > data/test_out.types
  
  4$ pardi -i data/decoys.mol2 -o data/still_decoys.mol2 \
             -d 's:@<TRIPOS>MOLECULE' -w 'cp %IN %OUT'
  
  4$ cat data/decoys.mol2 |
       parallel -n 1 --pipe --cat --recstart '@<TRIPOS>MOLECULE' \
         'cp {} {#}; cat {#}; rm {#}' > data/still_decoys.mol2
  
  5$ pardi -i data/decoys.mol2 -o data/decoys2.mol2 \
             -d b:10000 -w 'cp %IN %OUT' --preserve
  
  5$ cat data/decoys.mol2 |
       parallel -k --pipe --block 10k --recend '' --cat \
         'cat {} > {#}; cat {#}; rm {#}' > data/decoys2.mol2

https://github.com/UnixJunkie/pardi (Last checked: 2021-01)

Summary (see legend above):

- - - I4 -  - -

- - - - - M6

O1 - O3 - - - O7 O8 - -

E1 - - - - - -

- - - - - - - - -

- -

bthread takes around 1 sec per MB of output. The maximal output line length is 1073741759.

You cannot quote space in the command, so you cannot run composed commands like sh -c "echo a; echo b".

https://gitlab.com/netikras/bthread (Last checked: 2024-06)

Summary (see legend above):

I1 - - - - - I7

M1 - - - - M6

- O2 O3 - - O6 - x x O10

E1 - - - - - -

- - - - - - - - -

- -

EXAMPLES FROM simple_gpu_scheduler MANUAL

  1$ simple_gpu_scheduler --gpus 0 1 2 < gpu_commands.txt

  1$ parallel -j3 --shuf \
     CUDA_VISIBLE_DEVICES='{=1 $_=slot()-1 =} {=uq;=}' \
       < gpu_commands.txt

  2$ simple_hypersearch \
       "python3 train_dnn.py --lr {lr} --batch_size {bs}" \
       -p lr 0.001 0.0005 0.0001 -p bs 32 64 128 |
       simple_gpu_scheduler --gpus 0,1,2

  2$ parallel --header : --shuf -j3 -v \
       CUDA_VISIBLE_DEVICES='{=1 $_=slot()-1 =}' \
       python3 train_dnn.py --lr {lr} --batch_size {bs} \
       ::: lr 0.001 0.0005 0.0001 ::: bs 32 64 128

  3$ simple_hypersearch \
       "python3 train_dnn.py --lr {lr} --batch_size {bs}" \
       --n-samples 5 -p lr 0.001 0.0005 0.0001 -p bs 32 64 128 |
       simple_gpu_scheduler --gpus 0,1,2

  3$ parallel --header : --shuf \
       CUDA_VISIBLE_DEVICES='{=1 $_=slot()-1; seq()>5 and skip() =}' \
       python3 train_dnn.py --lr {lr} --batch_size {bs} \
       ::: lr 0.001 0.0005 0.0001 ::: bs 32 64 128

  4$ touch gpu.queue
     tail -f -n 0 gpu.queue | simple_gpu_scheduler --gpus 0,1,2 &
     echo "my_command_with | and stuff > logfile" >> gpu.queue

  4$ touch gpu.queue
     tail -f -n 0 gpu.queue |
       parallel -j3 CUDA_VISIBLE_DEVICES='{=1 $_=slot()-1 =} {=uq;=}' &
     # Needed to fill job slots once
     seq 3 | parallel echo true >> gpu.queue
     # Add jobs
     echo "my_command_with | and stuff > logfile" >> gpu.queue
     # Needed to flush output from completed jobs
     seq 3 | parallel echo true >> gpu.queue

https://github.com/ExpectationMax/simple_gpu_scheduler (Last checked: 2024-06)

parasweep is a Python module for facilitating parallel parameter sweeps.

A parasweep job will normally take a text file as input. The text file contains arguments for the job. Some of these arguments will be fixed and some of them will be changed by parasweep.

It does this by having a template file such as template.txt:

  Xval: {x}
  Yval: {y}
  FixedValue: 9
  # x with 2 decimals
  DecimalX: {x:.2f}
  TenX: ${x*10}
  RandomVal: {r}

and from this template it generates the file to be used by the job by replacing the replacement strings.

Being a Python module parasweep integrates tighter with Python than GNU parallel. You get the parameters directly in a Python data structure. With GNU parallel you can use the JSON or CSV output format to get something similar, but you would have to read the output.

parasweep has a filtering method to ignore parameter combinations you do not need.

Instead of calling the jobs directly, parasweep can use Python's Distributed Resource Management Application API to make jobs run with different cluster software.

GNU parallel --tmpl supports templates with replacement strings. Such as:

  Xval: {x}
  Yval: {y}
  FixedValue: 9
  # x with 2 decimals
  DecimalX: {=x $_=sprintf("%.2f",$_) =}
  TenX: {=x $_=$_*10 =}
  RandomVal: {=1 $_=rand() =}

that can be used like:

  parallel --header : --tmpl my.tmpl={#}.t myprog {#}.t \
    ::: x 1 2 3 ::: y 1 2 3

Filtering is supported as:

  parallel --filter '{1} > {2}' echo ::: 1 2 3 ::: 1 2 3

https://github.com/eviatarbach/parasweep (Last checked: 2021-01)

Summary (see legend above):

I1 I2 - - - - -

- - M3 - - M6

- O2 O3 - O5 O6 - O8 x O10

E1 - - - - - -

- - - - - - - - -

- -

parallel-bash is written in pure bash. It is really fast (overhead of ~0.05 ms/job compared to GNU parallel's 3-10 ms/job). So if your jobs are extremely short lived, and you can live with the quite limited command, this may be useful.

It works by making a queue for each process. Then the jobs are distributed to the queues in a round robin fashion. Finally the queues are started in parallel. This works fine, if you are lucky, but if not, all the long jobs may end up in the same queue, so you may see:

  $ printf "%b\n" 1 1 1 4 1 1 1 4 1 1 1 4 |
      time parallel -P4 sleep {}
  (7 seconds)
  $ printf "%b\n" 1 1 1 4 1 1 1 4 1 1 1 4 |
      time parallel-bash -p 4 -c sleep {}
  (12 seconds)

Because it uses bash lists, the total number of jobs is limited to 167000..265000 depending on your environment. You get a segmentation fault, when you reach the limit.

Ctrl-C does not stop spawning new jobs. Ctrl-Z does not suspend running jobs.

EXAMPLES FROM parallel-bash

  1$ some_input | parallel-bash -p 5 -c echo

  1$ some_input | parallel -j 5 echo

  2$ parallel-bash -p 5 -c echo < some_file

  2$ parallel -j 5 echo < some_file

  3$ parallel-bash -p 5 -c echo <<< 'some string'

  3$ parallel -j 5 -c echo <<< 'some string'

  4$ something | parallel-bash -p 5 -c echo {} {}

  4$ something | parallel -j 5 echo {} {}

https://github.com/Akianonymus/parallel-bash/ (Last checked: 2024-06)

Summary (see legend above):

I1 I2 - - - - -

- - M3 - - M6

- O2 O3 - O5 O6 - O8 x O10

E1 - - - - - -

- - - - - - - - -

- -

parallel-bash is written in pure bash. It is really fast (overhead of ~0.05 ms/job compared to GNU parallel's 3-10 ms/job). So if your jobs are extremely short lived, and you can live with the quite limited command, this may be useful.

It seems the number of jobs must be divisible by -p, so it sometimes does not run the jobs:

  # Does nothing
  $ seq 3 | parallel-bash -p 4 bash -c 'touch myfile-{}'

This should create myfile-1..3, but creates nothing.

It splits the input into queues. Each queue is of length -p. So this will make 250 queues and run all 250 processes in parallel:

  $ seq 1000 | parallel-bash -p 4 bash -c 'sleep {}'

This is quite different from parallel-bash(2021) where -p is the number of workers - similar to --jobs in GNU parallel.

In other words: parallel-bash does not quarantee that only 4 jobs will be run in parallel. This can overload your machine:

  # Warning: This will start 25000 processes - not just 4
  $ seq 100000 | parallel-bash -p 4 sleep {}

If you are unlucky all long jobs may end up in the same queue:

  $ printf "%b\n" 1 1 1 1 5 5 5 5 1 1 1 1 |
      time parallel -P4 sleep {}
  (7 seconds)
  $ printf "%b\n" 1 1 1 1 5 5 5 5 1 1 1 1 |
      time ./parallel-bash.bash -p 4 -c sleep {}
  (20 seconds)

Ctrl-C kills the jobs (as expected). Ctrl-Z does not suspend running jobs.

EXAMPLES FROM parallel-bash

  1$ main() { echo "${1}" ;}
     export -f main

  1$ printf "%b\n" {1..1000} | ./parallel-bash -p 10 main {}

  1$ printf "%b\n" {1..1000} | parallel -j 100 main {}

  2$ # Number of inputs must be divisible by 5
     some_input | parallel-bash -p 5 echo

  2$ some_input | parallel -j 5 echo

  3$ # Number of inputs must be divisible by 5
     parallel-bash -p 5 echo < some_file

  3$ parallel -j 5 echo < some_file

  4$ # Number of lines in 'some string' must be divisible by 5
     parallel-bash -p 5 echo <<< 'some string'

  4$ parallel -j 5 -c echo <<< 'some string'

  5$ something | parallel-bash -p 5 echo {}

  5$ something | parallel -j 5 echo {}

https://github.com/Akianonymus/parallel-bash/ (Last checked: 2024-06)

bash-concurrent is more an alternative to make than to GNU parallel. Its input is very similar to a Makefile, where jobs depend on other jobs.

It has a nice progress indicator where you can see which jobs completed successfully, which jobs are currently running, which jobs failed, and which jobs were skipped due to a depending job failed. The indicator does not deal well with resizing the window.

Output is cached in tempfiles on disk, but is only shown if there is an error, so it is not meant to be part of a UNIX pipeline. If bash-concurrent crashes these tempfiles are not removed.

It uses an O(n*n) algorithm, so if you have 1000 independent jobs it takes 22 seconds to start it.

https://github.com/themattrix/bash-concurrent (Last checked: 2024-06)

Summary (see legend above):

I1 - - - - - -

M1 - - - - M6

- O2 O3 - O5 O6 - x x O10

E1 - - - - - -

- - - - - - - - -

- -

spawn reads a full command line from stdin which it executes in parallel.

http://code.google.com/p/spawntool/ (Last checked: 2021-07)

Summary (see legend above):

- - - - - - -

M1 - - - - -

O1 - - - - - - x x O10

E1 - - - - - -

R1 R2 - - - R6 - - -

- -

go-pssh does ssh in parallel to multiple machines. It runs the same command on multiple machines similar to --nonall.

The hostnames must be given as IP-addresses (not as hostnames).

Output is sent to stdout (standard output) if command is successful, and to stderr (standard error) if the command fails.

EXAMPLES FROM go-pssh

  1$ go-pssh -l <ip>,<ip> -u <user> -p <port> -P <passwd> -c "<command>"

  1$ parallel -S 'sshpass -p <passwd> ssh -p <port> <user>@<ip>' \
       --nonall "<command>"

  2$ go-pssh scp -f host.txt -u <user> -p <port> -P <password> \
       -s /local/file_or_directory -d /remote/directory

  2$ parallel --nonall --slf host.txt \
       --basefile /local/file_or_directory/./ --wd /remote/directory
       --ssh 'sshpass -p <password> ssh -p <port> -l <user>' true

  3$ go-pssh scp -l <ip>,<ip> -u <user> -p <port> -P <password> \
       -s /local/file_or_directory -d /remote/directory

  3$ parallel --nonall -S <ip>,<ip> \
       --basefile /local/file_or_directory/./ --wd /remote/directory
       --ssh 'sshpass -p <password> ssh -p <port> -l <user>' true

https://github.com/xuchenCN/go-pssh (Last checked: 2021-07)

Summary (see legend above):

I1 I2 - - - - (I7)

- - M3 - - M6

- O2 O3 - O5 - - x x - O10

E1 - - E4 - - -

- - - - - - - - -

- -

go-parallel uses Go templates for replacement strings. Quite similar to the {= perl expr =} replacement string.

The basic replacement strings can be emulated by putting this into ~/.parallel/config:

  --rpl '{{.Input}} '
  --rpl '{{.Time}} use DateTime; $_= DateTime->from_epoch(time);'
  --rpl '{{.Start}} use DateTime; $_= DateTime->from_epoch($^T);'

Then you can do:

  seq 10 | parallel sleep {{.Input}}';' echo {{.Start}} {{.Time}}
  seq 10 | go-parallel -t 'bash -c "sleep {{.Input}}; echo \"{{.Start}}\" \"{{.Time}}\""'

If the input is too long (64K), you get no error:

   perl -e 'print "works."."x"x100' | parallel.go -t 'echo {{noExt .Input}} '
   perl -e 'print "fails."."x"x100_000_000' | parallel.go -t 'echo {{noExt .Input}} '

Special chars are quoted:

   echo '"&+<>' | go-parallel echo
   "&+<>

but not shell quoted when using replacement strings:

   echo '"&+<>' | go-parallel -t 'echo {{.Input}}'
   &#34;&amp;&#43;&lt;&gt;

EXAMPLES FROM go-parallel

  1$ go-parallel -a ./files.txt -t 'cp {{.Input}} {{.Input | dirname | dirname}}'

  1$ parallel -a ./files.txt cp {} '{= $_=::dirname(::dirname($_)) =}'

  2$ go-parallel -a ./files.txt -t 'mkdir -p {{.Input}} {{noExt .Input}}'

  2$ parallel -a ./files.txt echo mkdir -p {} {.}

  3$ go-parallel -a ./files.txt -t 'mkdir -p {{.Input}} {{.Input | basename | noExt}}'

  3$ parallel -a ./files.txt echo mkdir -p {} {/.}

  4$ time find ~/src/go -type f | go-parallel md5sum > /dev/null

  4$ time find ~/src/go -type f | parallel md5sum > /dev/null
     # Though you would probably do this instead:
     time find ~/src/go -type f | parallel -X md5sum > /dev/null

https://github.com/mylanconnolly/parallel (Last checked: 2024-06)

Summary (see legend above):

- - - I4 - - x

- - - - - M6

- O2 O3 - O5 O6 - x x - O10

E1 - - - - - -

- - - - - - - - -

- -

p is a tiny shell script. It can color output with some predefined colors, but is otherwise quite limited.

It maxes out at around 116000 jobs (probably due to limitations in Bash).

EXAMPLES FROM p

Some of the examples from p cannot be implemented 100% by GNU parallel: The coloring is a bit different, and GNU parallel cannot have --tag for some inputs and not for others.

The coloring done by GNU parallel is not exactly the same as p.

  1$ p -bc blue "ping 127.0.0.1" -uc red "ping 192.168.0.1" \
     -rc yellow "ping 192.168.1.1" -t example "ping example.com"

  1$ parallel --lb -j0 --color --tag ping \
     ::: 127.0.0.1 192.168.0.1 192.168.1.1 example.com

  2$ p "tail -f /var/log/httpd/access_log" \
     -bc red "tail -f /var/log/httpd/error_log"

  2$ cd /var/log/httpd;
     parallel --lb --color --tag tail -f ::: access_log error_log

  3$ p tail -f "some file" \& p tail -f "other file with space.txt"

  3$ parallel --lb tail -f ::: 'some file' "other file with space.txt"

  4$ p -t project1 "hg pull project1" -t project2 \
     "hg pull project2" -t project3 "hg pull project3"

  4$ parallel --lb hg pull ::: project{1..3}

https://github.com/rudymatela/evenmoreutils/blob/master/man/p.1.adoc (Last checked: 2022-04)

Summary (see legend above):

I1 - - - - - -

M1 - M3 - - M6

O1 - O3 O4 - - - x x -

E1 - - - - - -

- - - - - - - - -

- -

seneschal only starts the first job after reading the last job, and output from the first job is only printed after the last job finishes.

1 byte of output requites 3.5 bytes of RAM.

This makes it impossible to have a total output bigger than the virtual memory.

Even though output is kept in RAM outputing is quite slow: 30 MB/s.

Output larger than 4 GB causes random problems - it looks like a race condition.

This:

  echo 1 | seneschal  --prefix='yes `seq 1000`|head -c 1G' >/dev/null

takes 4100(!) CPU seconds to run on a 64C64T server, but only 140 CPU seconds on a 4C8T laptop. So it looks like seneschal wastes a lot of CPU time coordinating the CPUs.

Compare this to:

  echo 1 | time -v parallel -N0 'yes `seq 1000`|head -c 1G' >/dev/null

which takes 3-8 CPU seconds.

EXAMPLES FROM seneschal README.md

  1$ echo $REPOS | seneschal --prefix="cd {} && git pull"

  # If $REPOS is newline separated
  1$ echo "$REPOS" | parallel -k "cd {} && git pull"
  # If $REPOS is space separated
  1$ echo -n "$REPOS" | parallel -d' ' -k "cd {} && git pull"

  COMMANDS="pwd
  sleep 5 && echo boom
  echo Howdy
  whoami"
  
  2$ echo "$COMMANDS" | seneschal --debug
  
  2$ echo "$COMMANDS" | parallel -k -v
  
  3$ ls -1 | seneschal --prefix="pushd {}; git pull; popd;"
  
  3$ ls -1 | parallel -k "pushd {}; git pull; popd;"
  # Or if current dir also contains files:
  3$ parallel -k "pushd {}; git pull; popd;" ::: */

https://github.com/TheWizardTower/seneschal (Last checked: 2022-06)

Summary (see legend above):

x x x x x x x

- x x x x x

x O2 O3 O4 O5 O6 - x x O10

E1 - - E4 - - -

- - - - - - - - -

S1 S2

async works like sem.

EXAMPLES FROM async

  1$ S="/tmp/example_socket"

     async -s="$S" server --start

     for i in {1..20}; do
         # prints command output to stdout
         async -s="$S" cmd -- bash -c "sleep 1 && echo test $i"
     done
     
     # wait until all commands are finished
     async -s="$S" wait
     
  1$ S="example_id"

     # server not needed

     for i in {1..20}; do
         # prints command output to stdout
         sem --bg --id "$S" -j100% "sleep 1 && echo test $i"
     done
     
     # wait until all commands are finished
     sem --fg --id "$S" --wait
     
  2$ # configure the server to run four commands in parallel
     async -s="$S" server -j4
     
     mkdir "/tmp/ex_dir"
     for i in {21..40}; do
         # redirects command output to /tmp/ex_dir/file*
         async -s="$S" cmd -o "/tmp/ex_dir/file$i" -- \
           bash -c "sleep 1 && echo test $i"
     done
     
     async -s="$S" wait
     
     # stops server
     async -s="$S" server --stop

  2$ # starting server not needed

     mkdir "/tmp/ex_dir"
     for i in {21..40}; do
         # redirects command output to /tmp/ex_dir/file*
         sem --bg --id "$S" --results "/tmp/ex_dir/file$i{}" \
           "sleep 1 && echo test $i"
     done

     sem --fg --id "$S" --wait

     # there is no server to stop

https://github.com/ctbur/async (Last checked: 2023-01)

Summary (see legend above):

- - - I4 - - x

M1 - - - - M6

- - O3 - - - - x - -

E1 - E3 - E5 - -

- - - - - - - - -

- -

tandem runs full commands in parallel. It is made for starting a "server", running a job against the server, and when the job is done, the server is killed.

More generally: it kills all jobs when the first job completes - similar to '--halt now,done=1'.

tandem silently discards some output. It is unclear exactly when this happens. It looks like a race condition, because it varies for each run.

  $ tandem "seq 10000" | wc -l
  6731 <- This should always be 10002

EXAMPLES FROM Demo

  tandem \
    'php -S localhost:8000' \
    'esbuild src/*.ts --bundle --outdir=dist --watch' \
    'tailwind -i src/index.css -o dist/index.css --watch'

  # Emulate tandem's behaviour
  PARALLEL='--color --lb  --halt now,done=1 --tagstring '
  PARALLEL="$PARALLEL'"'{=s/ .*//; $_.=".".$app{$_}++;=}'"'"
  export PARALLEL

  parallel ::: \
    'php -S localhost:8000' \
    'esbuild src/*.ts --bundle --outdir=dist --watch' \
    'tailwind -i src/index.css -o dist/index.css --watch'

EXAMPLES FROM tandem -h

  # Emulate tandem's behaviour
  PARALLEL='--color --lb  --halt now,done=1 --tagstring '
  PARALLEL="$PARALLEL'"'{=s/ .*//; $_.=".".$app{$_}++;=}'"'"
  export PARALLEL

  1$ tandem 'sleep 5 && echo "hello"' 'sleep 2 && echo "world"'

  1$ parallel ::: 'sleep 5 && echo "hello"' 'sleep 2 && echo "world"'

  # '-t 0' fails. But '--timeout 0 works'
  2$ tandem --timeout 0 'sleep 5 && echo "hello"' \
       'sleep 2 && echo "world"'

  2$ parallel --timeout 0 ::: 'sleep 5 && echo "hello"' \
       'sleep 2 && echo "world"'

EXAMPLES FROM tandem's readme.md

  # Emulate tandem's behaviour
  PARALLEL='--color --lb  --halt now,done=1 --tagstring '
  PARALLEL="$PARALLEL'"'{=s/ .*//; $_.=".".$app{$_}++;=}'"'"
  export PARALLEL

  1$ tandem 'next dev' 'nodemon --quiet ./server.js'

  1$ parallel ::: 'next dev' 'nodemon --quiet ./server.js'

  2$ cat package.json
     {
       "scripts": {
         "dev:php": "...",
         "dev:js": "...",
         "dev:css": "..."
       }
     }

     tandem 'npm:dev:php' 'npm:dev:js' 'npm:dev:css'

  # GNU Parallel uses bash functions instead
  2$ cat package.sh
     dev:php() { ... ; }
     dev:js() { ... ; }
     dev:css() { ... ; }
     export -f dev:php dev:js dev:css

     . package.sh
     parallel ::: dev:php dev:js dev:css

  3$ tandem 'npm:dev:*'

  3$ compgen -A function | grep ^dev: | parallel

For usage in Makefiles, include a copy of GNU Parallel with your source using `parallel --embed`. This has the added benefit of also working if access to the internet is down or restricted.

https://github.com/rosszurowski/tandem (Last checked: 2023-01)

Summary (see legend above):

I1 I2 I3 - - - -

- - - - - M6

O1 O2 O3 - O5 O6 - x - O10

E1 - - E4 - - -

- - - - - - - - -

- -

rust-parallel has a goal of only using Rust. It seems it is impossible to call bash functions from the command line. You would need to put these in a script.

Calling a script that misses the shebang line (#! as first line) fails.

EXAMPLES FROM rust-parallel's README.md

  $ cat >./test <<EOL
  echo hi
  echo there
  echo how
  echo are
  echo you
  EOL
  
  1$ cat test | rust-parallel -j5
  
  1$ cat test | parallel -j5
  
  2$ cat test | rust-parallel -j1
  
  2$ cat test | parallel -j1
  
  3$ head -100 /usr/share/dict/words | rust-parallel md5 -s
  
  3$ head -100 /usr/share/dict/words | parallel md5 -s
  
  4$ find . -type f -print0 | rust-parallel -0 gzip -f -k
  
  4$ find . -type f -print0 | parallel -0 gzip -f -k
  
  5$ head -100 /usr/share/dict/words |
       awk '{printf "md5 -s %s\n", $1}' | rust-parallel
  
  5$ head -100 /usr/share/dict/words |
       awk '{printf "md5 -s %s\n", $1}' | parallel
  
  6$ head -100 /usr/share/dict/words | rust-parallel md5 -s |
       grep -i abba
  
  6$ head -100 /usr/share/dict/words | parallel md5 -s |
       grep -i abba

https://github.com/aaronriekenberg/rust-parallel (Last checked: 2023-01)

Summary (see legend above):

- I2 - - - - -

M1 - - - - M6

O1 - O3 - - - - x - -

E1 - - E4 - - -

- - - - - - - - -

- -

parallelium merges standard output (stdout) and standard error (stderr). The maximal output of a command is 8192 bytes. Bigger output makes parallelium go into an infinite loop.

In the input file for parallelium you can define a tag, so that you can select to run only these commands. A bit like a target in a Makefile.

Progress is printed on standard output (stdout) prepended with '#' with similar information as GNU parallel's --bar.

EXAMPLES

    $ cat testjobs.txt
    #tag common sleeps classA
    (sleep 4.495;echo "job 000")
    :
    (sleep 2.587;echo "job 016")
    
    #tag common sleeps classB
    (sleep 0.218;echo "job 017")
    :
    (sleep 2.269;echo "job 040")
    
    #tag common sleeps classC
    (sleep 2.586;echo "job 041")
    :
    (sleep 1.626;echo "job 099")
    
    #tag lasthalf, sleeps, classB
    (sleep 1.540;echo "job 100")
    :
    (sleep 2.001;echo "job 199")

    1$ parallelium -f testjobs.txt -l logdir -t classB,classC

    1$ cat testjobs.txt |
         parallel --plus --results logdir/testjobs.txt_{0#}.output \
           '{= if(/^#tag /) { @tag = split/,|\s+/ }
               (grep /^(classB|classC)$/, @tag) or skip =}'

https://github.com/beomagi/parallelium (Last checked: 2023-01)

Summary (see legend above):

I1 - - - - - I7

- - - - - -

- O2 O3 - O5 - - - - O10

E1 - - E4 - - -

- - - - - - - - -

- -

forkrun blocks if it receives fewer jobs than slots:

  echo | forkrun -p 2 echo

or when it gets some specific commands e.g.:

  f() { seq "$@" | pv -qL 3; }
  seq 10 | forkrun f

It is not clear why.

It is faster than GNU parallel (overhead: 1.2 ms/job vs 3 ms/job), but way slower than parallel-bash (0.059 ms/job).

Running jobs cannot be stopped by pressing CTRL-C.

-k is supposed to keep the order but fails on the MIX testing example below. If used with -k it caches output in RAM.

If forkrun is killed, it leaves temporary files in /tmp/.forkrun.* that has to be cleaned up manually.

EXAMPLES

  1$ time find ./ -type f |
       forkrun -l512 -- sha256sum 2>/dev/null | wc -l
  1$ time find ./ -type f |
       parallel -j28 -m -- sha256sum 2>/dev/null | wc -l

  2$ time find ./ -type f |
       forkrun -l512 -k -- sha256sum 2>/dev/null | wc -l
  2$ time find ./ -type f |
       parallel -j28 -k -m -- sha256sum 2>/dev/null | wc -l

https://github.com/jkool702/forkrun (Last checked: 2023-02)

Summary (see legend above):

I1 I2 - I4 - - -

M1 - - - - M6

O1 O2 O3 - O5 O6 - - - O10

E1 - - E4 - - -

- - - - - - - - -

- -

parallel-sh buffers in RAM. The buffering data takes O(n^1.5) time:

2MB=0.107s 4MB=0.175s 8MB=0.342s 16MB=0.766s 32MB=2.2s 64MB=6.7s 128MB=20s 256MB=64s 512MB=248s 1024MB=998s 2048MB=3756s

It limits the practical usability to jobs outputting < 256 MB. GNU parallel buffers on disk, yet is faster for jobs with outputs > 16 MB and is only limited by the free space in $TMPDIR.

parallel-sh can kill running jobs if a job fails (Similar to --halt now,fail=1).

EXAMPLES

  1$ parallel-sh "sleep 2 && echo first" "sleep 1 && echo second"

  1$ parallel ::: "sleep 2 && echo first" "sleep 1 && echo second"

  2$ cat /tmp/commands
     sleep 2 && echo first
     sleep 1 && echo second

  2$ parallel-sh -f /tmp/commands

  2$ parallel -a /tmp/commands

  3$ echo -e 'sleep 2 && echo first\nsleep 1 && echo second' |
       parallel-sh

  3$ echo -e 'sleep 2 && echo first\nsleep 1 && echo second' |
       parallel

https://github.com/thyrc/parallel-sh (Last checked: 2023-04)

Summary (see legend above):

- I2 - - - - I7

M1 - M3 - M5 M6

- O2 O3 - - O6 - O8 - O10

E1 - - - - - -

- - - - - - - - -

- -

bash-parallel is not as much a command as it is a shell script that you have to alter. It requires you to change the shell function process_job that runs the job, and set $MAX_POOL_SIZE to the number of jobs to run in parallel.

It is half as fast as GNU parallel for short jobs.

https://github.com/thilinaba/bash-parallel (Last checked: 2024-06)

Summary (see legend above): N/A

pash is quite different from GNU parallel. It is not a general parallelizer. It takes a shell script and analyses it and parallelizes parts of it by replacing the parts with commands that will give the same result.

This will replace sort with a command that does pretty much the same as parsort --parallel=8 (except somewhat slower):

  pa.sh --width 8 -c 'cat bigfile | sort'

However, even a simple change will confuse pash and you will get no parallelization:

  pa.sh --width 8 -c 'mysort() { sort; }; cat bigfile | mysort'
  pa.sh --width 8 -c 'cat bigfile | sort | md5sum'

From the source it seems pash only looks at: awk cat col comm cut diff grep head mkfifo mv rm sed seq sort tail tee tr uniq wc xargs

For pipelines where these commands are bottlenecks, it might be worth testing if pash is faster than GNU parallel.

pash does not respect $TMPDIR but always uses /tmp. If pash dies unexpectantly it does not clean up.

https://github.com/binpash/pash (Last checked: 2023-05)

Summary (see legend above):

I1 - - - - - -

M1 - - - - M6

- - O3 - - - - x x -

E1 - - - - - -

R1 - - - - R6 x x -

- -

korovkin-parallel prepends all lines with some info.

The output is colored with 6 color combinations, so job 1 and 7 will get the same color.

You can get similar output with:

  (echo ...) |
    parallel --color -j 10 --lb --tagstring \
      '[l:{#}:{=$_=sprintf("%7.03f",::now()-$^T)=} {=$_=hh_mm_ss($^T)=} {%}]'

Lines longer than 8192 chars are broken into lines shorter than 8192. korovkin-parallel loses the last char for lines exactly 8193 chars long.

Short lines from different jobs do not mix, but long lines do:

  fun() {
    perl -e '$a="'$1'"x1000000; for(1..'$2') { print $a };';
    echo;
  }
  export -f fun
  (echo fun a 100;echo fun b 100) | korovkin-parallel | tr -s abcdef
  # Compare to:
  (echo fun a 100;echo fun b 100) | parallel | tr -s abcdef

There should be only one line of a's and one line of b's.

Just like GNU parallel korovkin-parallel offers a master/slave model, so workers on other servers can do some of the tasks. But contrary to GNU parallel you must manually start workers on these servers. The communication is neither authenticated nor encrypted.

It caches output in RAM: a 1GB line uses ~2.5GB RAM

https://github.com/korovkin/parallel (Last checked: 2023-07)

Summary (see legend above):

I1 I2 - I4 - - I7

M1 - M3 M4 - M6

- O2 O3 - O5 O6 - O8 - O10

E1 - - E4 - - -

- - - - - - - - -

- -

xe has a peculiar limitation:

  echo /bin/echo | xe {} OK
  echo echo | xe /bin/{} fails

EXAMPLES

Compress all .c files in the current directory, using all CPU cores:

  1$ xe -a -j0 gzip -- *.c

  1$ parallel gzip ::: *.c

Remove all empty files, using lr(1):

  2$ lr -U -t 'size == 0' | xe -N0 rm

  2$ lr -U -t 'size == 0' | parallel -X rm

Convert .mp3 to .ogg, using all CPU cores:

  3$ xe -a -j0 -s 'ffmpeg -i "${1}" "${1%.mp3}.ogg"' -- *.mp3

  3$ parallel ffmpeg -i {} {.}.ogg ::: *.mp3

Same, using percent rules:

  4$ xe -a -j0 -p %.mp3 ffmpeg -i %.mp3 %.ogg -- *.mp3

  4$ parallel --rpl '% s/\.mp3// or skip' ffmpeg -i %.mp3 %.ogg ::: *.mp3

Similar, but hiding output of ffmpeg, instead showing spawned jobs:

  5$ xe -ap -j0 -vvq '%.{m4a,ogg,opus}' ffmpeg -y -i {} out/%.mp3 -- *

  5$ parallel -v --rpl '% s/\.(m4a|ogg|opus)// or skip' \
       ffmpeg -y -i {} out/%.mp3 '2>/dev/null' ::: *

  5$ parallel -v ffmpeg -y -i {} out/{.}.mp3 '2>/dev/null' ::: *

https://github.com/leahneukirchen/xe (Last checked: 2023-08)

Summary (see legend above):

- - - I4 - - -

M1 - M3 - - M6

- O2 O3 - O5 (O6) - x x O10

E1 - - - - - -

- - - - - - - - -

- -

sp has very few options.

It can either be used like:

  sp command {} option :: arg1 arg2 arg3

which is similar to:

  parallel command {} option ::: arg1 arg2 arg3

Or:

  sp command1 :: "command2 -option" :: "command3 foo bar"

which is similar to:

  parallel ::: command1 "command2 -option" "command3 foo bar"

sp deals badly with too many commands: This causes sp to run out of file handles and gives data loss.

For each command that fails, sp will print an error message on stderr (standard error).

You cannot used exported shell functions as commands.

EXAMPLES

  1$ sp echo {} :: 1 2 3

  1$ parallel echo {} ::: 1 2 3

  2$ sp echo {} {} :: 1 2 3

  2$ parallel echo {} {} :: 1 2 3

  3$ sp echo 1 :: echo 2 :: echo 3

  3$ parallel ::: 'echo 1' 'echo 2' 'echo 3'

  4$ sp a foo bar :: "b 'baz  bar'" :: c

  4$ parallel ::: 'a foo bar' "b 'baz  bar'" :: c

https://github.com/SergioBenitez/sp (Last checked: 2023-10)

Summary (see legend above):

- - - - - - -

- - - - - -

- O2 O3 x - O6 - x x ?O10

E1 - - - E5 - -

- - - - - - - - -

- -

repeater runs the same job repeatedly. In other words: It does not read arguments, thus is it an alternative for GNU parallel for only quite limited applications.

repeater has an overhead of around 0.23 ms/job. Compared to GNU parallel's 2-3 ms this is fast. Compared to bash-parallel's 0.05 ms/job it is slow.

Memory use and run time for large output

Output takes O(n^2) time for output of size n. 10 MB takes ~1 second, 30 MB takes ~7 seconds, 100 MB takes ~60 seconds, 300 MB takes ~480 seconds, 1000 MB takes ~10000 seconds.

100 MB of output takes around 1 GB of RAM.

    # Run time = 15 sec
    # Memory use = 20 MB
    # Output = 1 GB per job
    \time -v parallel -j1 seq ::: 120000000 120000000 >/dev/null

    # Run time = 4.7 sec
    # Memory use = 95 MB
    # Output = 8 MB per job
    \time -v repeater -w 1 -n 2 -reportFile ./run_output seq 1200000 >/dev/null

    # Run time = 42 sec
    # Memory use = 277 MB
    # Output = 27 MB per job
    \time -v repeater -w 1 -n 2 -reportFile ./run_output seq 3600000 >/dev/null

    # Run time = 530 sec
    # Memory use = 1000 MB
    # Output = 97 MB per job
    \time -v repeater -w 1 -n 2 -reportFile ./run_output seq 12000000 >/dev/null

    # Run time = 2h41m
    # Memory use = 8.6 GB
    # Output = 1 GB per job
    \time -v repeater -w 1 -n 2 -reportFile ./run_output seq 120000000 >/dev/null

For even just moderate sized outputs GNU parallel will be faster and use less memory.

EXAMPLES

  1$ repeater -n 100 -w 10 -reportFile ./run_output
       -output REPORT_FILE -progress BOTH curl example.com

  1$ seq 100 | parallel --joblog run.log --eta curl example.com > output

  2$ repeater -n 100 -increment -progress HIDDEN -reportFile foo
       echo "this is increment: " INC
  2$ seq 100 | parallel echo {}
  2$ seq 100 | parallel echo '{= $_ = ++$myvar =}'

https://github.com/baalimago/repeater (Last checked: 2023-12)

Summary (see legend above):

I1 - - - - - I7

- - - - - M6

O1 - O3 O4 O5 - O7 - - -

E1 - - E4 - - -

- - - - - - - - -

- -

parallelize runs the full line as a command. If the command is not found, there is no warning.

The output at most ~1000000 lines/s. If the lines are short this is quite slow. The lines can at most be 2047999 bytes long. Longer lines cause segfault.

EXAMPLES

  simple.dat:
  
    sleep 5
    foo
    cat alire.toml
    loc src/parallelize.adb
    sh loc src/*.ad?
  
  1$ bin/parallelize -v <simple.dat

  1$ parallel <simple.dat

https://github.com/simonjwright/parallelize (Last checked: 2024-04)

https://github.com/justanhduc/task-spooler

https://manpages.ubuntu.com/manpages/xenial/man1/tsp.1.html

https://www.npmjs.com/package/concurrently

http://code.google.com/p/push/ (cannot compile)

https://github.com/krashanoff/parallel

https://github.com/Nukesor/pueue

https://arxiv.org/pdf/2012.15443.pdf KumQuat

https://github.com/JeiKeiLim/simple_distribute_job

https://github.com/reggi/pkgrun - not obvious how to use

https://github.com/benoror/better-npm-run - not obvious how to use

https://github.com/bahmutov/with-package

https://github.com/flesler/parallel

https://github.com/Julian/Verge

https://vicerveza.homeunix.net/~viric/soft/ts/

https://github.com/chapmanjacobd/que

Output from 2 jobs should not mix. If the output is not used, this does not matter; but if the output is used then it is important that you do not get half a line from one job followed by half a line from another job.

If the tool does not buffer, output will most likely mix now and then.

This test stresses whether output mixes.

  #!/bin/bash

  paralleltool="parallel -j 30"

  cat <<-'EOF' > mycommand
  #!/bin/bash

  # If a, b, c, d, e, and f mix: Very bad
  perl -e 'print STDOUT "a"x3000_000," "'
  perl -e 'print STDERR "b"x3000_000," "'
  perl -e 'print STDOUT "c"x3000_000," "'
  perl -e 'print STDERR "d"x3000_000," "'
  perl -e 'print STDOUT "e"x3000_000," "'
  perl -e 'print STDERR "f"x3000_000," "'
  echo "stdout line 1 of id $@"
  echo "stderr line 1 of id $@" >&2
  perl -e 'print STDOUT "A"x3000_000," "'
  perl -e 'print STDERR "B"x3000_000," "'
  perl -e 'print STDOUT "C"x3000_000," "'
  perl -e 'print STDERR "D"x3000_000," "'
  perl -e 'print STDOUT "E"x3000_000," "'
  perl -e 'print STDERR "F"x3000_000," "'
  echo "stdout line 2 of id $@"
  echo "stderr line 2 of id $@" >&2
  EOF
  chmod +x mycommand

  # Run 30 jobs in parallel
  seq 30 |
    $paralleltool -j 30  ./mycommand > >(tr -s a-zA-Z) 2> >(tr -s a-zA-Z >&2)

  # 'a c e' and 'b d f' should always stay together
  # For each job there be 2 lines of standard output and standard error
  # They should not be interleaved with other id's

Output from stdout and stderr should not be merged, but kept separated.

This test shows whether stdout is mixed with stderr.

  #!/bin/bash

  paralleltool="parallel -j0"

  cat <<-EOF > mycommand
  #!/bin/bash

  echo stdout
  echo stderr >&2
  echo stdout
  echo stderr >&2
  EOF
  chmod +x mycommand

  # Run one job
  echo |
    $paralleltool ./mycommand > stdout 2> stderr
  cat stdout
  cat stderr

Some tools cache output in RAM. This makes them extremely slow if the output is bigger than physical memory and crash if the output is bigger than the virtual memory.

  #!/bin/bash

  paralleltool="parallel -j0"

  cat <<'EOF' > mycommand
  #!/bin/bash

  # Generate 1 GB output
  yes "`perl -e 'print \"c\"x30_000'`" | head -c 1G
  EOF
  chmod +x mycommand

  # Run 20 jobs in parallel
  # Adjust 20 to be > physical RAM and < free space on /tmp
  seq 20 | time $paralleltool ./mycommand | wc -c

If caching is done on disk, the disk can run full during the run. Not all programs discover this. GNU Parallel discovers it, if it stays full for at least 2 seconds.

  #!/bin/bash

  paralleltool="parallel -j0"

  # This should be a dir with less than 100 GB free space
  smalldisk=/tmp/shm/parallel
  
  TMPDIR="$smalldisk"
  export TMPDIR
  
  max_output() {
      # Force worst case scenario:
      # Make GNU Parallel only check once per second
      sleep 10
      # Generate 100 GB to fill $TMPDIR
      # Adjust if /tmp is bigger than 100 GB
      yes | head -c 100G >$TMPDIR/$$
      # Generate 10 MB output that will not be buffered
      # due to full disk
      perl -e 'print "X"x10_000_000' | head -c 10M
      echo This part is missing from incomplete output
      sleep 2
      rm $TMPDIR/$$
      echo Final output
  }
  
  export -f max_output
  seq 10 | $paralleltool max_output | tr -s X

Some tools do not clean up tmp files if they are killed. If the tool buffers on disk, they may not clean up, if they are killed.

  #!/bin/bash

  paralleltool=parallel

  ls /tmp >/tmp/before
  seq 10 | $paralleltool sleep &
  pid=$!
  # Give the tool time to start up
  sleep 1
  # Kill it without giving it a chance to cleanup
  kill -9 $!
  # Should be empty: No files should be left behind
  diff <(ls /tmp) /tmp/before

It is not uncommon for users to create files like:

  My brother's 12" *** record  (costs $$$).jpg

Some tools break on this.

  #!/bin/bash

  paralleltool=parallel

  touch "My brother's 12\" *** record  (costs \$\$\$).jpg"
  ls My*jpg | $paralleltool ls -l

Some tools require you to wrap composed commands into bash -c.

  echo bar | $paralleltool echo foo';' echo {}

Some tools can only insert the argument once.

  echo bar | $paralleltool echo {} foo {}

Some tools limit the length of the input lines artificially with no good reason. GNU parallel does not:

  perl -e 'print "foo."."x"x100_000_000' | parallel echo {.}

GNU parallel limits the command to run to 128 KB due to execve(1):

  perl -e 'print "x"x131_000' | parallel echo {} | wc

Some tools become very slow if output lines have many words.

  #!/bin/bash

  paralleltool=parallel

  cat <<-EOF > mycommand
  #!/bin/bash

  # 10 MB of lines with 1000 words
  yes "`seq 1000`" | head -c 10M
  EOF
  chmod +x mycommand

  # Run 30 jobs in parallel
  seq 30 | time $paralleltool -j0 ./mycommand > /dev/null

  #!/bin/bash
  
  paralleltool="parallel -j0"

  wcc() {
      parallel --recend '' --block 100M --pipe 'LC_ALL=C wc' |
        datamash -W sum 1 sum 2 sum 3
  }

  cat <<-EOF > mycommand
  #!/bin/bash
  
  perl -e '\$a="a"x1000_000; for(1..5000) { print \$a }'
  EOF
  chmod +x mycommand
  
  # Run 1 job
  seq 1 | $paralleltool ./mycommand | wcc

Permission is granted to copy, distribute and/or modify this documentation under the terms of the GNU Free Documentation License, Version 1.3 or any later version published by the Free Software Foundation; with no Invariant Sections, with no Front-Cover Texts, and with no Back-Cover Texts.  A copy of the license is included in the file LICENSES/GFDL-1.3-or-later.txt.

You are free:

to Share

to copy, distribute and transmit the work

to Remix

to adapt the work

Under the following conditions:

Attribution

You must attribute the work in the manner specified by the author or licensor (but not in any way that suggests that they endorse you or your use of the work).

Share Alike

If you alter, transform, or build upon this work, you may distribute the resulting work only under the same, similar or a compatible license.

With the understanding that:

Waiver

Any of the above conditions can be waived if you get permission from the copyright holder.

Public Domain

Where the work or any of its elements is in the public domain under applicable law, that status is in no way affected by the license.

Other Rights

In no way are any of the following rights affected by the license:

• Your fair dealing or fair use rights, or other applicable copyright exceptions and limitations;
• The author's moral rights;
• Rights other persons may have either in the work itself or in how the work is used, such as publicity or privacy rights.

Notice

For any reuse or distribution, you must make clear to others the license terms of this work.

A copy of the full license is included in the file as LICENCES/CC-BY-SA-4.0.txt