HPCBIOS_05-05: Common Queue NamesΒΆ


Common queue names will allow users to understand HPCBIOS compatible resources faster, select appropriate queues and, move workflows across sites without related issues.

  • BC Policy: HPCBIOS_05-05
  • Date of Policy: 2012-12-15

The goal of this policy is to institute a core namespace of batch queues with the same names at each of the systems participating in a Baseline Configuration. The queue names were chosen to align with collective experience from multiple HPC sites. In addition, each system or site is explicitly allowed to offer customized batch support and service levels for specialized workloads or projects of each own. Systems MAY have more queues on as needed basis and SHOULD extend this policy in a backward-compatible way.

The core set of common queue names will allow users to more easily move their jobs and scripts among systems. This improves user productivity by requiring less time and effort to support many different scripts and to learn a different queue environment at each site.

The Compute queue names defined by HPCBIOS and their respective priority classes are:

Queue Name Priority Class Cost factor Policy SLA Description Specifications to meet (ie. minimum maxima)
urgent Urgent 1 MAY Approved by Site as needed, for fulfilling site or project objectives
interactive Interactive 1 MAY For interactive jobs maxsize @ 4 nodes; max. walltime @ 2 hours
debug Debug 0 SHOULD Time/Resource limited for debug purposes max. walltime @ 30 minutes
gpu Standard 1 MAY For approved projects needing GPUs Normal queue for GPU resources
batch Standard 1 MUST For approved projects - GENERIC maxsize @ 50% of resources ; max. walltime @ 24hours
transfer Standard 1 MAY Data stage-in/stage-out jobs used for jobs heavy in I/O, meant for data transfers
gpu-low Low Priority 0 MAY For approved projects needing GPUs No allocation subtraction, possibly best-effort
batch-low Low Priority 0 MAY For approved projects No allocation subtraction, possibly best-effort

The above list of queues, scaling among urgent and batch-low, denotes different level of priorities, ranging from the highest to lowest.

A site still maintains the right to choose which users have access to which queues and at which times, so essentially compliance with this policy is quite flexible.

Potential sources of information for further future upgrades of this policy include:

Kindly notify -if this policy is inadequate for your work- both your local site technical representative & HPCBIOS maintainers.