Comet

Main events

Comet is lead by academics: Dr. ? and Dr. ?. The hardware has been bought by Lenovo, which also handles all the hardware and sofware maintinence. RSE-HPC team will freely provide support and user training, but the research software engineering team can provide more hands on support if there is budget for it.

The comet wiki is editable by anyone (although the edits are reviewd before approval).

The HPC portal provides a point of access to most of the managerial options and technical information about the HPC services.

Costs

HPC can be used for free, as long as the jobs are linked to a project that has been approved by the HPC committee. The amount of resources allocated to free projects is limited; larger projects should be founded. This ensures that they have access to more resources, from more performant GPU nodes to an higher threshold for the number of jobs that can be run in parallel.

Scheduler

The scheduler will prioritize small jobs over large jobs, in order to fit as many jobs as possible in the queue.

Research Data Warehouse (RDW)

Research Data Warehouse (RDW) is file storage for research, hosted on our campus datacentre. It is possible to view the files from both Windows and Linux, permissions on the files are set using your campus login so it’s essential that you are logged in as the correct campus user in order to see your files.

Requesting new HPC projects

Anyone can request a new HPC project, but

• the project funding status (unfunded/funded) must be set, but it can be changed later;
• all people associated with the project must have passed the HPC driving test;
• at least two people associated with the project must be Newcastle University staff.

Software modules

Comet comes with a large number of software modules pre-installed. If there is a request for a new module

Loading diagram...

Job checkpoints are supported!

Running jobs

All jobs must necessarely specify

• the account (the name of the project) to which the job belongs;
• the kind of node (partition) to which the job is to be submitter.

Partitions

• short_(free|paid) - 1 hour
• default_(free|paid) - 24 hours
• low_latency_paid - 24 hours, for funded projects only, low latency nodes
• gpu_(free|paid) - 24 hours, paid get access to better GPUs (L40A vs H100)
• interactive - 1 hour, for funded projects only, interactive nodes

Open OnDemand

Open OnDemand (OOD) is a web-based interface to the HPC cluster, which allows users to submit jobs, manage files, and access interactive sessions without needing to use the command line. It just requires a web browser and a campus login, and it can be accessed from anywhere.