Getting started on clusters
Simple Linux Utility for Resource Management (SLURM)
SLURM is an open source, fault-tolerant, and highly scalable cluster management and job scheduling system for large and small Linux clusters. SLURM requires no kernel modifications for its operation and is relatively self-contained. As a cluster workload manager, SLURM has three key functions. First, it allocates exclusive and/or non-exclusive access to resources (compute nodes) to users for some duration of time so they can perform work. Second, it provides a framework for starting, executing, and monitoring work (normally a parallel job) on the set of allocated nodes. Finally, it arbitrates contention for resources by managing a queue of pending work. Optional plugins can be used for accounting, advanced accounting, gang scheduling (time sharing for parallel jobs), backfill scheduling, topology optimized resource selection, resource limits by user or bank account, and sophisticated multifactor job prioritization algorithms. This document presumes zero prior knowledge of cluster computing. If instead you're an intermediate user (e.g. you have an account and have run a few jobs before but need a reminder) the table of contents is your friend.
SLURM has a centralized manager,
slurmctld , to monitor resources and work. There may also be a backup manager to assume those responsibilities in the event of failure. Each compute server (node) has a
slurmd , daemon, which can be compared to a remote shell: it waits for work, executes that work, returns status, and waits for more work. The
slurmd , daemons provide fault-tolerant hierarchical communications. There is an optional
slurmdbd (Slurm DataBase Daemon) which can be used to record accounting information for multiple Slurm-managed clusters in a single database. User tools include
srun to initiate jobs,
scancel to terminate queued or running jobs,
sinfo to report system status,
squeue to report the status of jobs, and
sacct to get information about jobs and job steps that are running or have completed. The
sview commands graphically reports system and job status including network topology. There is an administrative tool
scontrol available to monitor and/or modify configuration and state information on the cluster. The administrative tool used to manage the database is
sacctmgr . It can be used to identify the clusters, valid users, valid bank accounts, etc. APIs are available for all functions.
This document gives you all the information you need to choose a system, log in to a cluster/phat node, write a script, submit it via qsub to the scheduler, and find the output. As such, these notes cover hardware and software. (If you're a sysadmin, you may be interested in this page instead.)
- Get a cluster account. You can email admin at cs dot earlham dot edu or a current CS faculty member to get started. Your user account will grant access to all the servers below, and you will have a home directory at
~usernamethat you can access when you connect to any of them.
- Note: if you have a CS account, you will use the same username and password for your cluster account.
- Connect through a terminal via ssh to
email@example.com. If you intend to work with these machines a lot, you should also configure your ssh keys.
Cluster systems to choose from
The cluster dot earlham dot edu domain consists of clusters (a collection of physical servers linked through a switch to perform high-performance computing tasks with distributed memory) and jumbo servers (nee "phat nodes"; a system comprising one physical server with a high ratio of disk+RAM to CPU, good for jobs demanding shared memory).
Our current machines are:
- whedon: newest cluster; 8 compute nodes; Torque-only pending an OS upgrade
- layout: cluster; 4 compute nodes, pre-whedon, features NVIDIA GPGPU's and multiple CUDA options
- lovelace: newest jumbo server
- pollock: jumbo server, older than lovelace but well-tested and featuring the most available disk space
To get to, e.g., whedon, from hopper, run
If you're still not sure, click here for more detailed notes.
Cluster software bundle
The cluster dot earlham dot edu servers all run a supported CentOS version.
All these servers (unless otherwise noted) also feature the following software:
- Slurm (scheduler): submit a job with
sbatch jobname.sbatch, delete it with
scancel jobID. Running a job has its own doc section below.
- Environment modules: run
module availto see available software modules and
module load modulenameto load one; you may load modules in bash scripts and qsub jobs as well.
The default shell on all these servers is bash.
The default Python version on all these servers is Python 2.x, but all have at least one Python 3 module with a collection of widely-used scientific computing libraries.
Slurm is our batch scheduler.
You can check that it's working by running:
srun -l hostname
You can submit a job in a script with the following:
Here's an example of a batch file:
#!/bin/sh #SBATCH --time=1 #SBATCH --job-name hello-world #SBATCH --nodes=1 #SBATCH -c 1 # ask for one core #SBATCH --mail-type=BEGIN,END,FAIL #SBATCH --firstname.lastname@example.org echo "queue/partition is `echo $SLURM_JOB_PARTITION`" echo "running on `echo $SLURM_JOB_NODELIST`" echo "work directory is `echo $SLURM_SUBMIT_DIR`" /bin/hostname srun -l /bin/hostname sleep 10 srun -l /bin/pwd
Interactive and command line interfaces also exist. After submitting a job slurm captures anything written to stdout and stderr by the programs and when the job completes puts it in a file called slurm-nnn.out (where nnn is the job number) in the directory where you ran sbatch. Use more to view it when you are looking for error messages, output file locations, etc.
If you are used to using
qpeek, you can instead just run
tail -f jobXYZ.out or
tail -f jobXYZ.err.
There's some more CPU management information here.
Conversion from Torque to Slurm
To submit a job to PBS, you'll need to write a shell script wrapper around it and submit it through qsub on your system of choice. For example (change the specific options):
||run/submit a batch job|
||show jobs currently in the queue|
||cancel a job|
||show nodes in the cluster|
||the queue/partition you are in|
||there's no equivalent of the nodes file but there is an environment variable that stores that information|
||working directory from which the command was run|
#!/usr/bin/bash #SBATCH --job-name hello-world #SBATCH --nodes=5 #SBATCH --mail-type=BEGIN,END,FAIL #SBATCH --email@example.com echo "queue is `echo $SLURM_JOB_PARTITION`" echo "running on `echo $SLURM_JOB_NODELIST`" echo "work directory is `echo $SLURM_SUBMIT_DIR`" srun -l echo "hello world!"
Some useful commands
Man pages exist for all Slurm daemons, commands, and API functions. The command option
--help also provides a brief summary of options. Note that the command options are all case sensitive.
sacct is used to report job or job step accounting information about active or completed jobs.
salloc is used to allocate resources for a job in real time. Typically this is used to allocate resources and spawn a shell. The shell is then used to execute srun commands to launch parallel tasks.
sattach is used to attach standard input, output, and error plus signal capabilities to a currently running job or job step. One can attach to and detach from jobs multiple times.
sbatch is used to submit a job script for later execution. The script will typically contain one or more
srun commands to launch parallel tasks.
sbcast is used to transfer a file from local disk to local disk on the nodes allocated to a job. This can be used to effectively use diskless compute nodes or provide improved performance relative to a shared file system.
scancel is used to cancel a pending or running job or job step. It can also be used to send an arbitrary signal to all processes associated with a running job or job step.
scontrol is the administrative tool used to view and/or modify SLURM state. Note that many
scontrol commands can only be executed as user root.
sinfo reports the state of partitions and nodes managed by SLURM. It has a wide variety of filtering, sorting, and formatting options.
sprio is used to display a detailed view of the components affecting a job's priority.
squeue reports the state of jobs or job steps. It has a wide variety of filtering, sorting, and formatting options. By default, it reports the running jobs in priority order and then the pending jobs in priority order.
srun is used to submit a job for execution or initiate job steps in real time.
srun has a wide variety of options to specify resource requirements, including: minimum and maximum node count, processor count, specific nodes to use or not use, and specific node characteristics (so much memory, disk space, certain required features, etc.). A job can contain multiple job steps executing sequentially or in parallel on independent or shared resources within the job's node allocation.
sshare displays detailed information about
fairshare usage on the cluster. Note that this is only viable when using the priority/multifactor plugin.
sstat is used to get information about the resources utilized by a running job or job step.
strigger is used to set, get or view event triggers. Event triggers include things such as nodes going down or jobs approaching their time limit.
sview is a graphical user interface to get and update state information for jobs, partitions, and nodes managed by SLURM.
Before Slurm we used Torque and its associated software, including qsub. This is now deprecated and should not be used on the Earlham CS cluster systems.