lunduniversity.lu.se

LUNARC

Lund University

Denna sida på svenska This page in English

Memory consumption will effect Aurora queuing priorities

2018-12-13

When more memory than the default memory per core is requested, more resources are blocked for other users than the core count. Recently there has been a large increase in jobs that use extra memory and a number of groups getting much more than their allocated share of resources, resulting in a larger queue and longer waiting times than usual. To restore a fair use of the resources for everybody, memory requirements are being taken into account into the calculation of usage as seen through the command projinfo, which also determines the queuing priorities. The amount of memory corresponding to the memory per core on a node is counted equal to one core and if the total memory request is larger than the number of requested cores, the memory request will determine the usage instead. 

Note that the priority is lowered not only when the allocation has been spent, but also when a project has running jobs using more than twice the number of cores or equivalent memory needed to achieve the allocation when running continuously for 30 days.

Most users will not notice the change other than that the length of the queuing time will decrease. Users who have been using more memory per core than the default amount will notice that their 30-day cumulative usage has increased to reflect their actual use of the system.

Please contact LUNARC support in case of comments or questions.


Page Manager:

Calendar