Speed up extraction od tar.bz2 files using bash

Shell Programming and Scripting
1

The below bash will untar each tar.bz2 folder in the directory, then remove the tar.bz2 .

Each of the tar.bz2 folders ranges from 40-75GB and currently takes ~2 hours to extract. Is there a way to speed up the extraction process?

I am using a xeon processor with 12 cores. Thank you :).

for i in /home/cmccabe/Desktop/NGS/API/*.tar.bz2; do 
    tar -xvjf "$i" -C /home/cmccabe/Desktop/NGS/API
rm $i
done
2

At 10 megabytes per second it sounds like there's some room for improvement. But you can't go too crazy or you'll just slow your disk down to uselessness.

This requires the BASH shell, mostly for the ability to do wait "$ONEPARTICULARTHREAD" instead of wait #for everything

#!/bin/bash

maxproc=2 # Max number of threads.  Suggest 2, or 3 at most
i=0

# Count files
set -- /home/cmccabe/Desktop/NGS/API/*.tar.bz2
FILES="$#"

# Blank $1 $2 ...
set --

let i=1
for FILE in /home/cmccabe/Desktop/NGS/API/*.tar.bz2
do
        printf "(%2d/%2d)\tProcessing %s\n" "$i" "$FILES" "$FILE"
        let i=i+1

        tar -xvjf "$FILE" -C /home/cmccabe/Desktop/NGS/API >/dev/null &

        # Turn $1=pida $2=pidb $3=pidc $4=pidd, into
        #      $1=pida $2=pidb $3=pidc $4=pidd $5=pide
        set -- "$@" $!

        # Shift removes $1 and moves the rest down, so you get
        # $1=pidb $2=pidc $3=pidd $3=pide
        # $# is the number of arguments.
        if [ "$#" -ge $maxproc ]  ; then wait "$1" ; shift; fi
done
1 Like
3

Hi.

I think parallel can help you:

NAME
       parallel - build and execute shell command lines from standard input in
       parallel
...

Some details on parallel:

parallel        build and execute shell command lines from standard in... (man)
Path    : /usr/bin/parallel
Version : 20130922
Length  : 6224 lines
Type    : Perl script, ASCII text executable
Shebang : #!/usr/bin/env perl
Repo    : Debian 8.7 (jessie) 
Home    : https://www.gnu.org/software/parallel/ (pm)
Modules : (for perl codes)
 IPC::Open3     1.16
 POSIX  1.38_03
 Symbol 1.07
 CGI::File::Temp        4.09
 File::Path     2.09
 Getopt::Long   2.42
 strict 1.08
 strict 1.08
 FileHandle     2.02
 POSIX  1.38_03parallel        build and execute shell command lines from standard in... (man)

Some help at:

       You can also watch the intro video for a quick introduction:
       http://tinyogg.com/watch/TORaR/ http://tinyogg.com/watch/hfxKj/ and
       http://tinyogg.com/watch/YQuXd/ or
       http://www.youtube.com/playlist?list=PL284C9FF2488BC6D1

Best wishes ... cheers, drl

1 Like
4

Can I just check that this files are local and not NFS mounted. If they are remote, then you are dependant on the network too, along with a dollop of memory to work on the file. You will also lose any caching that could help you.

If the files are local, then ignore me.

Robin

1 Like
5

Yes, if the files are NFS mounted, my attempt or "parallel" will both hurt, not help!

1 Like
6

The tar.bz2 folders are local. I tried parallel using:

pbzip2 -dvc folder.tar.bz2 | tar x
pbzip2 -v -d -k -m10500 folder.tar.bz2 | tar x

Those did execute but were really no faster. The second command uses the max allowed 20MB to decompress.

This code is extremely fast but seems to extract partial files within each tar.bz2 .

Contents of folder.tar.bz2

file1.bam -20GB
file2.bam -25GB
file3.bam -19GB
file1.vcf - 10MB
file2.vcf - 8MB
file3.vcf -10MB
file1.bam.bai - 1MB
file2.bam.bai - 1MB
file3.bam.bai - 1 MB

#!/bin/bash

maxproc=2 # Max number of threads.  Suggest 2, or 3 at most
i=0

# Count files
set -- /home/cmccabe/Desktop/NGS/API/*.tar.bz2
FILES="$#"

# Blank $1 $2 ...
set --

let i=1
for FILE in /home/cmccabe/Desktop/NGS/API/*.tar.bz2
do
        printf "(%2d/%2d)\tProcessing %s\n" "$i" "$FILES" "$FILE"
        let i=i+1

        tar -xvjf "$FILE" -C /home/cmccabe/Desktop/NGS/API >/dev/null &

        # Turn $1=pida $2=pidb $3=pidc $4=pidd, into
        #      $1=pida $2=pidb $3=pidc $4=pidd $5=pide
        set -- "$@" $!

        # Shift removes $1 and moves the rest down, so you get
        # $1=pidb $2=pidc $3=pidd $3=pide
        # $# is the number of arguments.
        if [ "$#" -ge $maxproc ]  ; then wait "$1" ; shift; fi
done

After the code executes:

file1.bam -15MB
file3.vcf -1MB
file1.bam.bai - 1MB

There may not always be 9 files in each folder, but the file types will always be .bam and .vcf and .bam.bai . Thank you :).

7

The command drl suggested was not pbzip but in fact parallel. Instead of extracting multiple partial files from one tar, you can get several tars extracting at once.

Which is what my code is for, actually.

I neglected one line at the end. It shouldn't have mattered, but if the code did manage to quit while the children were running, its possible it killed them instead of waiting. So:

#!/bin/bash

maxproc=2 # Max number of threads.  Suggest 2, or 3 at most
i=0

# Count files
set -- /home/cmccabe/Desktop/NGS/API/*.tar.bz2
FILES="$#"

# Blank $1 $2 ...
set --

let i=1
for FILE in /home/cmccabe/Desktop/NGS/API/*.tar.bz2
do
        printf "(%2d/%2d)\tProcessing %s\n" "$i" "$FILES" "$FILE"
        let i=i+1

        tar -xvjf "$FILE" -C /home/cmccabe/Desktop/NGS/API >/dev/null &

        # Turn $1=pida $2=pidb $3=pidc $4=pidd, into
        #      $1=pida $2=pidb $3=pidc $4=pidd $5=pide
        set -- "$@" $!

        # Shift removes $1 and moves the rest down, so you get
        # $1=pidb $2=pidc $3=pidd $3=pide
        # $# is the number of arguments.
        if [ "$#" -ge $maxproc ]  ; then wait "$1" ; shift; fi
done

wait
1 Like
8

It was the wait , thank you very much it speeds up the untar process by 50% :).

1 Like