HPC Cluster Overview

High-Performance Computing at CHTC

The CHTC high-performance computing (HPC) cluster provides dedicated support for large, singular computations that use specialized software (i.e. MPI) to achieve internal parallelization of work across multiple servers of dozens to hundreds of cores.

Is high-performance computing right for me? Only computational work that fits that above description is appropriate for the HPC Cluster. Computational work that can complete on a single node in less than a few days will be best supported by our larger high-throughput computing (HTC) system (which also includes specialized hardware for extreme memory, GPUs, and other cases). For more information, please see Our Approach.

To get access to the HPC Cluster, please complete our New User Consultation Form. After your request is received, a Research Computing Facilitator will follow up to discuss the computational needs of your research and connect you with computing resources (including non-CHTC services) that best fit your needs.

HPC Cluster User Policies

See our User Policies and Expectations for details on general CHTC and HPC cluster policies.

HPC Hardware and Configuration

The HPC Cluster consists of two login nodes and many compute (aka execute) nodes. All users log in at a login node, and all user files on the shared file sytem are accessible on all nodes. Additionally, all nodes are tightly networked (200 Gbit/s Infiniband) so they can work together as a single "supercomputer", depending on the number of CPUs you specify.

Operating System and Software

All nodes in the HPC Cluster are running CentOS 8 Stream Linux.

The SLURM scheduler version is 22.05.6.

To see more details of other software on the cluster, see the HPC Software page.

Login Nodes

The login node for the cluster is: spark-login.chtc.wisc.edu

For more details on logging in, see the “Connecting to CHTC” guide linked above.

Execute Nodes and Partitions

Only execute nodes will be used for performing your computational work. The execute nodes are organized into several "partitions", including the shared, pre, and int partitions which are available to all HPC users as well as research group specific partitions that consist of researcher-owned hardware and which all HPC users can access on a backfill capacity via the pre partition (more details below).

• shared compute nodes each have 64 or 128 cores and 512 GB of RAM.
  Jobs submitted to this partition can request and use up to 7 days of running time.

• int consists of two compute nodes is intended for short and immediate interactive testing on a single node (up to 16 CPUs, 64 GB RAM). Jobs submitted to this partition can run for up to 4 hours.

• pre (i.e. pre-emptable) is an under-layed partition encompassing all HPC Cluster nodes and is intended for more immediate turn-around of shorter, smaller, and/or interactive sessions requiring more than the 4 hour time limit of the int partition. Jobs submitted to pre are run as back-fill on any idle nodes, including researcher-owned compute nodes, meaning these jobs may be pre-empted by higher priority jobs. By default, pre-empted jobs will be re-queued (to run again) if they were submitted with an sbatch script.

Fair Share Allocation

To promote fair access to HPC computing resources, all users are limited to 10 concurrently running jobs (if you need to queue more, please get in touch). Additionally, users are restricted to a total of 720 cores across all running jobs (core limits do not apply on research group partitions of more than 720 cores).

When determining which order to run jobs, the following policies are applies, in order or significance to job priority determinations:

A. User priority decreases as the user accumulates hours of CPU time over the last 21 days, across all queues. This “fair-share” policy means that users who have run many/larger jobs in the near-past will have a lower priority, and users with little recent activity will see their waiting jobs start sooner. (The cluster does not have a strict “first-in-first-out” queue policy.)

B. Job priority increases with job wait time. After the history-based user priority calculation in (A), the next most important factor for each job’s priority is the amount of time that each job has already waited in the queue. For all the jobs of a single user, these jobs will most closely follow a “first-in-first-out” policy.

C. Job priority increases with job size, in cores. This least important factor slightly favors larger jobs, so that the scheduler can take advantage when large numbers of newly-available nodes happen to become available (requiring less wasted time to deliberately drain nodes for larger jobs). So, among a user’s jobs submitted at roughly the same time, a larger job may run first, if the number of nodes necessary for the larger job is already available.

Data Storage and Management

Data space in the HPC Cluster filesystem is not backed-up and should be treated as temporary by users. Only files necessary for actively-running jobs should be kept on the filesystem, and files should be removed from the cluster when jobs complete. A primary copy of any essential files (e.g. software, submit files, input) should be kept in an alternate, non-CHTC storage location.

Each user will receive two primary data storage locations:

1. /home/username with an initial disk quota of 30GB and 250,000 items. Your home directory is meant to be used for files you use repeatedly, like submit file templates, source code, software installations, and reference data files.

2. /scratch/username with an initial disk quota of 100GB and 250,000 items. Jobs should always be submitted and run out of /scratch. It is the space for all working data, including individual job inputs, job outputs, and job log/stderr/stdout files.

What about /software?

If you are installing software meant to be shared within a group, we can create a dedicated folder for you in the /software space email us (chtc@cs.wisc.edu) if this is you!

To check how many files and directories you have in your /home or /scratch directory see the instructions below.

Changes to quotas for either of these locations are available upon request per our Request a Quota Change guide. If you don't know how many files your installation creates, because it's more than the current items quota, simply indicate that in your request.

CHTC Staff reserve the right to remove any significant amounts of data on the HPC Cluster in our efforts to maintain filesystem performance for all users.

Local scratch space is available on each execute node in /local/$USER. This space is NOT automatically cleaned out, so if you use this space, be sure to remove the files before the end of your job script or interactive session.

Tools for managing home and software space

You can use the command get_quotas to see what disk and items quotas are currently set for a given directory path. This command will also let you see how much disk is in use and how many items are present in a directory:

[username@hpclogin1 ~]$ get_quotas /home/username /scratch/username

Alternatively, the ncdu command can also be used to see how many files and directories are contained in a given path:

[username@hpclogin1 ~]$ ncdu /home/username
[username@hpclogin1 ~]$ ncdu /scratch/username

When ncdu has finished running, the output will give you a total file count and allow you to navigate between subdirectories for even more details. Type q when you're ready to exit the output viewer. More info here: https://lintut.com/ncdu-check-disk-usage/