Skip to content

Accounts, Access and Policies

Accounts

Request an account

Please create a ticket in ServiceNow in which you mention: 1. Your name and P-number 2. The name of your department 3. Cost center / Kostenplaats

Address the ticket to the ServiceNow group HPC support.

Costs

Costs for usage are calculated per month.

Overview TBA

Access

Policies

Terms of Use

Use of the HELIOS cluster is only allowed in compliance with all the usual rules and conditions set by the Amsterdam UMC for ICT systems. In general, the conditions apply that are published by the Central ICT department: Aansluitvoorwaarden Dienst ICT and Kader Informatiebeveiliging – Authenticatie: gebruikersnamen en wachtwoorden.

In case of careless handling of privacy sensitive data, or in case of conscious violation of applicable rules, the possibility exists that affected accounts are closed for a shorter or longer period of time.

Personal account

It is not allowed to share your personal Amsterdam UMC account with other users. Each username has an owner, who is responsible for the careful use of this account.

Guidelines

The HELIOS cluster is used by many people at the same time. Unfortunately, this also means that your activity can negatively impact other peoples' work.

Please read the Best Practices carefully when you start working on the cluster

Best practices

In short

  1. Close unused applications.
  2. Store local files only in your personal folder /net/beegfs/scratch/<p-number>. Keep your home folder as clean as possible to prevent login problems.
  3. Delete or backup files that are not part of an ongoing analysis. The scratch folder is only for temporary storage. The network shares can be used for archiving.
  4. Use slurm for large or lengthy jobs, or when using GPU. Usage of compute resources is monitored, limited by quota and costs are accounted to your cost center. Don't oversize slurm requests.
  5. When you encounter any problems on the server, first try to find a solution on this wiki. If that doesn't help, contact the support team.

Data Storage

Do not store (research) data in your local home (~) folder, but in /net/beegfs/scratch/<p-number>. Keep local storage on the server as lean as possible; Store4ever, L-disk and R-disk are for long-term storage and backups. Data on the server (i.e. /scratch/... and /home/...) are not backed up.

Make sure to cleanup your folders on the server and network disks at the end of your project, internship or contract. Archive important data to store4ever, the R-disk or L-disk. Archive source code in gitlab. Personal data on the server (i.e. /home and /scratch) will automatically be removed without further notice when your AMC or Amsterdam UMC account is disabled or removed by the Dienst ICT or HR.

Data transfer is only allowed for data that you are allowed to process (i.e. DTA, DPIA etc in place). Do not active network connections that are semi-permanent or enable (ssh) tunneling to other systems. Accounts with suspicious or unsafe activity will be disabled immediately (either by your admins or Dienst ICT).

Always use the job manager Slurm

On the system the job manager Slurm is used to manage the computational jobs that users want to run and schedule these jobs in such a way that the system is optimally used without overloading it. This way, the system can process everybody's job requests as fast as possible.

It is therefore essential that you always use the job manager, both for interactive (test and development) work and for scheduled (background) jobs. Otherwise, the system will perform sub-optimal for all users and may even crash in some cases. Be aware that the system will abort a running application without further notice as soon as the system detects that the total amount of unmanaged resources reaches a preset limit. This page describes all information about slurm. Regular (non-slurm) login sessions are rather limited in available resources. Use the show-my-session-limits command to view your limits.

Priority

The computing power of the central processing units (CPUs) is shared between all active processes. The Linux operating system tries to distribute the available computing time among running processes using very short portions of time.

Parallel

Some applications can spawn multiple processes or threads from a single command, often as a default without you specifying it. Although this can reduce the required computation time significantly, it can cause unintentional high CPU loads on the server.

In rare cases, this could bring down the whole system. Therefore, keep an eye on your jobs by checking the actual CPU and memory usage of your application. The easiest way to do this is with the command line tool top.

There is no 'one size fits all' solution for this problem, so it depends on the application how built-in parallel processing can be limited. Matlab and Mathematica have configurable setting for the maximum number of parallel workers. The taskset utility can be used to limit the number of cores/threads that will be used by an application. However, if you (intentionally) would like to make use of multithreading or parallel processing, you can specify this when running the application as a slurm job.

GPU

The cluster has multiple GPU cards that can be used with e.g. the CUDA library. Some cards are divided into smaller subsystems with less internal memory and calculation units (MIG mode). This way more applications can use the available GPU resources at the same time without interfering with each other.

Be sure to minimize the use of GPU resources by choosing the smallest possible configuration that allows the application to work well. The use of GPU resources, like other resources, is registered per user in the slurm account.

Be Tidy

Software licenses and disk space are limited resources that are shared between users. Therefore, only use what you really need. Close unused applications and remove unused data from the system.

Disk usage

In some cases it may be useful to create a temporary copy of your data to the scratch disk. The scratch disk is a fast disk that is directly connected to the server and is much faster than the network shares. However, the scratch disk has limited capacity. To prevent problems for other users, you should keep track of the disk space that is used in your ~/my-scratch folder (disk quota apply). Delete or backup unused files. The tar and zip command can be used to compress and archive large datasets.