Parallel: A Command-line CPU Load Balancer Written in Rust
This is an attempt at recreating the functionality of GNU Parallel in Rust under a MIT license. The end goal will be to support much of the functionality of GNU Parallel
and then to extend the functionality further for the next generation of command-line utilities written in Rust.
Benchmark Comparison to GNU Parallel
Here are some benchmarks from an i5-2410M laptop running Gentoo.
GNU Parallel:
Default Options
real 0m5.728s
user 0m2.960s
sys 0m1.310s
Executed with the --ungroup
Option
real 0m4.801s
user 0m2.070s
sys 0m1.290s
Rust Parallel:
Default Options
The default options are the slowest options, with all features enabled.
real 0m1.198s
user 0m0.130s
sys 0m0.550s
Executed with the --no-shell
option
A significant amount of overhead is caused by executing commands within the platform's preferred shell. On Unix
systems, that shell is sh
, whereas on Windows it is cmd
. Disabling shell executing is a good idea if your
command is simple and doesn't require chaining multiple commands.
real 0m0.559s
user 0m0.084s
sys 0m0.372s
Executed with the --no-shell
and --ungroup
Option
This will achieve utmost optimization at the cost of not having the standard output and error printed in order.
real 0m0.575s
user 0m0.060s
sys 0m0.450s
Syntax Examples
The following syntax is supported:
|
Options
In addition to the command syntax, there are also some options that you can use to configure the load balancer:
- -h, --help: Prints the manual for the application (recommended to pipe it to
less
). - -j, --jobs: Defines the number of jobs/threads to run in parallel.
- -u, --ungroup: By default, stdout/stderr buffers are grouped in the order that they are received.
- -n, --no-shell: Disables executing commands within the platform's shell for a performance boost.
- -v, --verbose: Prints information about running processes.
- --num-cpu-cores: Prints the number of CPU cores in the system and exits.
Available syntax options for the placeholders values are:
- {}: Each occurrence will be replaced with the name of the input.
- {.}: Each occurrence will be replaced with the input, with the extension removed.
- {/}: Each occurrence will be replaced with the base name of the input.
- {/.}: Each occurrence will be replaced with the base name of the input, with the extension removed.
- {//}: Each occurrence will be replaced with the directory name of the input.
- {%}: Each occurrence will be replaced with the slot number.
- {#}: Each occurrence will be replaced with the job number.
- {#^}: Each occurrence will be replaced with the total number of jobs.
Useful Examples
Transcoding FLAC music to Opus
ffmpeg is a highly useful application for converting music and videos. However, audio transcoding is limited to a a single core. If you have a large FLAC archive and you wanted to compress it into the efficient Opus codec, it would take forever with the fastest processor to complete, unless you were to take advantage of all cores in your CPU.
Transcoding Videos to VP9
VP9 has one glaring flaw in regards to encoding: it can only use about three cores at any given point in time. If you have an eight core processor and a dozen or more episodes of a TV series to transcode, you can use the parallel program to run three jobs at the same time, provided you also have enough memory for that.
vp9_params="-c:v libvpx-vp9 -tile-columns 6 -frame-parallel 1 -rc_lookahead 25 -threads 4 -speed 1 -b:v 0 -crf 18"
opus_params="-c:a libopus -b:a 128k"
How It Works
There are a lot of commands that will take an input and then consume an entire CPU core as it processes the input. However, sometimes you have dozens, hundreds, or even thousands of files that you want to process. The standard solution would be to construct a for loop and run your jobs serially one at a time. However, this would take forever with processes that only make use of a single core. Another solution is to construct the same for loop but to have your shell run it in the background. The problem with that solution is that if there are a lot of inputs to process, you will end locking up your system and crashing your jobs due to OOM (out of memory) errors.
A complicated setup that I have seen people perform is to create as many separate lists or directories as they have CPU cores, and then manually spinning up a terminal and copying and pasting the same for loop into each one. The issue with this approach is that it takes a lot of time to set this up, and because some tasks finish much sooner than others, you may end up with several cores sitting and waiting because they've completed all of their assigned inputs while other cores are busy with many more tasks left to perform.
Instead of processing files using a for loop, you can use a load balancer like parallel
to distribute jobs evenly
to every core in the system, which will only pass new values when a core has finished it's task. This has the benefit
that you can process inputs chronologically, and because some inputs may finish sooner than others, you can ensure
that every core has a job to process at any given point in time. Not to mention, it's about as easy to write as a
for loop:
# This is a for loop
for; do ; done
# This is a parallel version of that for loop
Installation Instructions
There are a number of methods that you can use to install the application. I provide binary packages for AMD64 systems that are available for download:
Ubuntu
Everyone Else
Compiling From Source
All of the dependencies are vendored locally, so it is possible to build the packages without Internet access.
First Method
If you would like to install the latest release directly to ~/.cargo/bin
using the official method.
Second Method
If you would like to install the latest git release:
Third Method
If you would like to install it system-wide.