Running a parallel job (alternative episode)

Last updated on 2026-09-15 | Edit this page

Estimated time: 90 minutes

ERROR 

Error in `find_config()`:
! Could not find lesson configuration in any known location.

Overview

Questions

  • What is the difference between array jobs and MPI?
  • What benefits arise from parallel execution?
  • What are the limits of gains from execution in parallel?

Objectives

  • Distinguish between job arrays and MPI
  • Download prime generator program
  • Prepare a job submission script for the parallel executable.
  • Launch jobs with parallel execution.
  • Record and summarize the timing and accuracy of jobs.
  • Describe the relationship between job parallelism and performance.

In the previous episode we mentioned the use of the Message Passing Interface (MPI) to accomplish parallelisation. While array jobs allow us to launch several instances of the same program, but with different data, across several nodes, MPI allows a single task to be distributed over several CPU cores.

It is thus possible to use array jobs in conjunction with MPI.

Callout

What is MPI?

The Message Passing Interface is a set of tools which allow multiple tasks running simultaneously to communicate with each other. Typically, a single executable is run multiple times, possibly on different machines, and the MPI tools are used to inform each instance of the executable about its sibling processes, and which instance it is. MPI also provides tools to allow communication between instances to coordinate work, exchange information about elements of the task, or to transfer data. An MPI instance typically has its own copy of all the local variables.

In this episode we will use two small programs, written in C, to calculate the number of primes found between two given numbers. One of the programs calculates prime and by using MPI spreads the job over several CPU cores while the other program doesn’t. After running both these programs one can compare their output to see the difference in efficiency.

If you disconnected, log back in to the cluster.

BASH 

 ssh @

Ideally the code for this episode should be pre-compiled and made available for students to download. We have found that expecting students to write or even compile code causes information overload and confusion.

The code and scripts to compile can be downloaded from https://github.com/NewcastleRSE-Training/HPC_Training_Example_Jobs. After compiling the two versions of the program, copy it to a place where students can copy or download it from.

The binaries of the two programs will be very small so the fact that there would be duplication if all the students copy the binaries to their own working directories should not really matter. In doing it this way, students will also get the opportunity to submit a job where the program they are using is in their local directory (rather than loading a module).

Only do this if pre-compiled binaries of the programs have not been made available to you.

Requirements

  • GCC
  • OpenMPI

Steps

  • Download code
  • Compile code
  • Copy binaries to home directory

If you disconnected, log back in to the cluster.

BASH 

 ssh @

Clone the repository

BASH 

 git clone https://github.com/NewcastleRSE-Training/HPC_Training_Example_Jobs.git

Compile the code.

BASH 

 cd HPC_Training_Example/c
 module load GCC OpenMPI
 ./compile.sh

OUTPUT 

Compiling primes.c function...
Compiling single process version...
Creating executable binary...
-rwxr-x--- 1 username group 17472 Jan 23 11:14 single_gcc

Compiling primes.c function...
Compiling single process version...
Creating executable binary...
-rwxr-x--- 1 username group 7176 Jan 23 11:14 single_aocc

Compiling primes.c function...
Compiling MPI multi-process version...
Creating executable binary...
-rwxr-x--- 1 username group 17096 Jan 23 11:14 multi

Move (or copy) the binaries to your home directory

Copying the programs into your home directory

Make sure you are in your home directory.

BASH 

 cd ~

You will need to amend the from-directory in the instruction below if you did not compile the code yourself according to the above challenge:

BASH 

 mkdir primes
 cd primes
 cp ~/HPC_Training_Example_Jobs/c/multi ~/HPC_Training_Example_Jobs/c/single_gcc .

Help!

Many command-line programs include a “help” message. Try it with single_gcc:

BASH 

 ./single_gcc

OUTPUT 

You must enter two positive numbers in the range 1 - 2^32

This message doesn’t tell us much about what the program does, but it does tell us that we need to provide two numbers that specify the beginning and the end of a range that lies between 1 and 2^32.

Callout

The time command

You will notice in the batch scripts that we will be creating we will be using the time command before the name of the program. For example:

BASH 

 time ./single_gcc

OUTPUT 

You must enter two positive numbers in the range 2 - 2^32

real	0m0.005s
user	0m0.000s
sys	0m0.002s

The very first line of the output is the output of the program we want to run, i.e. single_gcc. After that time returns three times. real is wall clock time. If you ran a stopwatch, that is how long it would have taken. The user time is the amount of CPU time it has taken. sys is kernel/system call time. That is the time the code spent doing things that were not part of your code, but essential stuff like interrupts, time the kernel spent setting up processes and memory.

Running the Job on a Compute Node

Create a submission file, requesting one task on a single node, then launch it.

ERROR 

Error in `snippets()`:
! snippets() called before configuration was loaded.

Use the status commands to check whether your job is running and when it ends:

BASH 

  --me

Use ls to locate the output file. The -t flag sorts in reverse-chronological order: newest first. What was the output?

Read the Job Output

The cluster output should be written to a file in the folder you launched the job from. For example,

BASH 

 ls -t

OUTPUT 

slurm-1177272.out  job_single.sh  job_multi.sh  single_gcc  multi

BASH 

 cat slurm-1177272.out

OUTPUT 

Starting single process primes calculation (2 - 10000000)
=====================
main: Calculating primes in the range 2 - 10000000
primeCount: Calculating primes 2 - 10000000
primeCount: Found 664579 primes
main: Found a total of 664579 primes

real	0m34.476s
user	0m34.247s
sys	0m0.002s
=====================
Primes calculation complete

While MPI-aware executables can generally be run as stand-alone programs, in order for them to run in parallel they must use an MPI run-time environment, which is a specific implementation of the MPI standard. To activate the MPI environment, the program should be started via a command such as mpiexec (or mpirun, or srun, etc. depending on the MPI run-time you need to use), which will ensure that the appropriate run-time support for parallelism is included.

Running the Parallel Job

The program multi uses the Message Passing Interface (MPI) for parallelism. – this is a common tool on HPC systems.

Callout

MPI Runtime Arguments

On their own, commands such as mpiexec can take many arguments specifying how many machines will participate in the execution, and you might need these if you would like to run an MPI program on your own (for example, on your laptop). In the context of a queuing system, however, it is frequently the case that MPI run-time will obtain the necessary parameters from the queuing system, by examining the environment variables set when the job is launched.

Let’s modify the job script to request more cores and use the MPI run-time.

ERROR 

Error in `snippets()`:
! snippets() called before configuration was loaded.
Challenge

Is it 16× faster?

The parallel job received 16× more processors than the serial job: does that mean it finished in 1/16th of the time?

The parallel job did take less time: 3.493s is better than 34.476s! But it is only a 9.87× improvement, not 16×.

Look at the job output:

  • While “process 0” did serial work, processes 1 through 3 did their parallel work.
  • While process 0 caught up on its parallel work, the rest did nothing at all.

Process 0 always has to finish its serial task before it can start on the parallel work. This sets a lower limit on the amount of time this job will take, no matter how many cores you throw at it.

This is the basic principle behind [Amdahl’s Law][amdahl], which is one way of predicting improvements in execution time for a fixed workload that can be subdivided and run in parallel to some extent.

In an HPC environment, we try to reduce the execution time for all types of jobs, and MPI is an extremely common way to combine dozens, hundreds, or thousands of CPUs into solving a single problem. To learn more about parallelization, see the parallel novice lesson lesson.

Key Points
  • Parallel programming allows applications to take advantage of parallel hardware.
  • The queuing system facilitates executing parallel tasks.
  • Performance improvements from parallel execution do not scale linearly.