Checking cluster status

This page shows the current queues for StarExec. The queue all.q is available for everyone to use. The other queues serve special functions or are reserved for exclusive use by a community (for instance, for competitions). Compute nodes are moved from one queue to another depending on need.

Note that only community leaders can request reservation of a queue.

You can select a queue to see jobs running on that queue as well as pending (enqueued but not yet executing) job pairs for that queue, or a node to see currently executing job pairs on that node.

Scheduling

When several users are attempting to use the same queue at the same time, pairs are scheduled to balance compute time among each user as evenly as possible. This is done by tracking the compute time used by each user and preferentially adding pairs for users that have taken the least runtime. Whenever a job pair is added to a queue, the pair's wallclock timeout is used as an estimation of the pair's runtime until the pair finishes, at which point the actual wallclock runtime of the pair is used. Note that this means that setting wallclock timeouts that are substantially higher than necessary may cause jobs to run more slowly in the short term.