General Conditions of Use
The use of ARCCA HPC facilities is subject to the general Cardiff University's computing policies, for full details please see Information Services IT regulations. When requesting an account on ARCCA systems all users agree to abide by these terms and conditions (as indicated on the account request form).
Never let anyone else use your account or disclose your password to friends or colleagues. In the event of these rules being breached your account will be immediately suspended.
Acceptable Usage Policy for Raven (HPC Cluster)
This policy is designed to enable a fair distribution of the resources over the entire user community whist retaining sufficient flexibility to meet short term demands from researchers. The policy is subject to revision by the ARCCA Operational and User Group to ensure the communities needs are being met whilst getting the maximum performance from the technology. Amendments will be based on current usage profiles and data collected from the system reporting logs.
ARCCA's plan is to become a sustainable facility and will ultimately need to recover all costs incurred by capital depreciation and the running costs of the centre (both support staff and infrastructure). The model used is the Major Research Facility (MRF) approach to full economic costing (fEC).Therefore ARCCA must have a clearly defined usage policy with service level objectives (SLO) to define the level of service provision for paying users. The ARCCA Operational and User Group are in the process of agreeing these definitions - these will be available online soon. It is acknowledged by this group that whilst ARCCA needs to recover costs it is essential to allow users with limited funding access to the machine - this is achieved through a University pump-prime basis, which has 30% of the system available to non-paying users.
In the interim, to ensure fair share, the following regulations have been agreed as standard usage policy for Raven:
- Users must not run serial applications interactively on the login nodes. This significantly impacts on the response time of the cluster and consequently on other users. PBS Pro can be used to reserve compute nodes to enable interactive jobs to be run. Please see the Raven user guide for more information.
- All computational jobs run on Raven must be submitted via the PBS Pro job scheduler. This will enable resources to be sensibly allocated and their usage tracked.
- Users must not use /tmp for even the temporary storage of large data files, as the filling up of the file-system may cause severe problems to the service. Please use /home as files left in /tmp are subject to summary deletion.
- /scratch should only be used to store working copies of data. Files older than on 60 days on the /scratch partition may be deleted. Please move important files back to the /home partition or to another system.
- Currently the maximum runtime for parallel jobs is 72 hours. Please contact us if this limitation causes you major concerns. If your jobs require less than 72 hours, please use the walltime flag in the job submission script to allocate a maximum run time for your calculation.
- Users should have no more than 20 jobs in the scheduler at any one time. Users who persistently submit large numbers of jobs may have their accounts suspended. There is a maximum number of jobs (10) any one user can have running in the scheduler. Please contact us if you need to submit a large number of jobs over a short period of time so we can make the necessary arrangements.
- Users are reminded that it is their responsibility to ensure that important data is adequately backed up. At the moment, ARCCA cannot guarantee that data stored on the system is resilient since it is not backed up as part of the default service.
- Users are requested to acknowledge the use of ARCCA facilities in their research publications and that they provide a brief summary of their research which might be used for publicity purposes (subject to approval from the researcher).
- The use of ARCCA facilities constitutes an agreement to provide ARCCA with details of any publications concerning research conducted using ARCCA resources. ARCCA request that researchers inform us of any research papers, presentations or posters which used our facilities to generate results to enable us to monitor the research outputs from the facility.
This policy will be amended periodically in consultation with the ARCCA Operational & User Group.
Acceptable Usage Policy for Condor
- New users will submit their jobs from a bank of dedicated submit servers.
- Users must not use /tmp for even the temporary storage of large data files as the filling up of the file system may cause severe problems to the service. Please use /home as files left in /tmp are subject to summary deletion.
- Maximum practical run time per job is 5 hours therefore you should aim to divide your work up into multiple jobs each of which takes no longer than 5 hours to run on a typical desktop computer which, as of this writing, is a computer with a 3.0GHz Dual-Core processor and 2GB of memory. The reason for this is to minimise energy consumption and therefore environmental impact.
- Maximum practical data size per job is 50 megabytes input or output therefore you should aim to divide your input data up into multiple chunks each of which is no larger than 50 megabytes and generate no more than 50 megabytes of output data per job. The reason for this is to minimise energy consumption and therefore environmental impact.
- Users should submit no more than 5000 jobs at any one time.
Please contact us if you are unable to divide your work up based on these limitations as it may be the case that we can find a way around them.