Meridian High-Performance Computing Cluster

Internal allocation

All Veyra Institute researchers and graduate students receive a baseline allocation of 2,000 node-hours per quarter, renewed automatically. Additional allocation is available through a competitive application process reviewed quarterly by the Research Computing Committee. Applications require a brief technical justification (one page) and an estimate of expected outputs.

Research groups with active external grants may apply for project-specific allocations that run for the duration of the grant. These are not drawn from the baseline pool and are not subject to quarterly caps.

External allocation

External users book compute time at 44 cr/hr per node through the Facilities Portal (booking code HPC-MER). A minimum booking of 4 node-hours applies; there is no maximum per booking, but allocations above 512 nodes require prior notification to the HPC team. External bookings are billed in arrears against the registered account.

Academic users from accredited institutions receive the standard 15% external user discount. Frame agreements for sustained external users are available for clients expecting to use more than 50,000 node-hours per year.

Fair-share scheduling

SLURM uses fair-share scheduling with a 7-day decay half-life. Users who have consumed less than their fair share receive priority; users who have over-consumed are deprioritised. The system converges to fair usage within approximately 48 hours of an imbalance.

A dedicated partition (priority) is available for time-sensitive jobs. Priority partition time is limited to 20% of each user's quarterly allocation and incurs a 1.5× billing multiplier for internal users (external users pay the standard rate regardless of partition).

Job limits

Default per-user limits: 512 concurrent nodes, 7-day walltime maximum. Jobs requiring more than 512 nodes or longer walltimes are accommodated via a reservation request submitted to the HPC team with at least 72 hours advance notice. Maintenance windows (one per quarter, typically Sunday 00:00–08:00) are published on the HPC status page six weeks in advance.

Data management

Home directories (100 GB quota) are backed up nightly. Scratch space (/scratch) is purged after 30 days and is not backed up. Project directories (allocation-based quota) are retained for the duration of the allocation plus 90 days. Long-term archiving to the tape library is available via an HSM policy on request.