Scheduling

Showing 44

Enable option for hardcoding GPU memory request like its available for CPU memory request

At this moment IBM offers features of hardcoding CPU memory. We have implemented this feature in our cluster with esub scripting. Similarly we want to have same feature available for GPU memmory (gmem) to be hardcoded. I have raised ticket to our ...

over 1 year ago in Spectrum LSF / Scheduling 2 Not under consideration

1 VOTE

Correct on Design Flaw in Session Scheduler

Currently the Session Manager and Service Director operate in the same CGROUP on the first execution host which causes a race condition that if a task on the first execution host pushes the memory use above the memory limit (aka mem-eater process)...

11 months ago in Spectrum LSF / Scheduling 1 Not under consideration

2 VOTE

bjobs -o reserved_mem or bhosts -o reserved_mem

To see how much memory is reserved on a host, we need to parse the bhosts -l <hostname> command, and it doesn't give us info about which job is reserving how much memory. I'd either like to see one of the following: bhosts -o 'jobid reserved...

almost 2 years ago in Spectrum LSF / Scheduling 1 Future consideration

4 VOTE

Set a limit for service class

It would be useful to set a limit in lsb.resources for Serive Class "SLA" in order give some minimum guarantee with SLA and set a maximium number of resorucese usable in this SLA.It would be excellent to indicate the limit with the number of nodes...

about 4 years ago in Spectrum LSF / Scheduling 1 Not under consideration

1 VOTE

lsfproxyd - Persistent blocking rules

LSFProxyd is a new solution to help scheduler handle queries effectively. This RFE is to enable persistent blocksing rules for heavy query hitters. Same has been discussed with John Welch in a separate discussion.

about 1 year ago in Spectrum LSF / Scheduling 1 Future consideration

1 VOTE

Cleanup user details with bqueues -lr output if they don't have running/done jobs in system

Currently, we can find thousands of entries for all users even though they have ZERO STARTED jobs as listed in bqueues -lr output. For some of larger farms, these entries are close to 7M records and bqueues -lr takes long time to return the data. ...

about 1 year ago in Spectrum LSF / Scheduling 2 Planned for future release

3 VOTE

Enhance GSLA to enable in-queue preemption

The Guaranteed SLA system allows sharing of Resources defined by ServiceClasses. We would like to use that to allow two or more groups to pool and share compute slots, wherein they typically unused slots would be available to all of the groups. Th...

almost 4 years ago in Spectrum LSF / Scheduling 1 Not under consideration

2 VOTE

Reporting both host-based pending reason and main pending reason

"LSF does not support reporting both host-based pending reason and main pending reason (about license)." Let me provide a portion of the bsub statement from our case to exemplify the customer's dilemma. bsub -R "rusage[mem=3000.00:duration=1m,md_r...

over 2 years ago in Spectrum LSF / Scheduling 1 Not under consideration

1 VOTE

Affinity support for windows compute servers

We have a mixed Linux/Windows cluster. Unfortunately the affinity string in the resource requirement is not supported on Windows. This leads to several problems: 1. jobs with affinity settings will not be dispatched to windows nodes 2. less perfor...

over 1 year ago in Spectrum LSF / Scheduling 1 Future consideration

1 VOTE

Use all applicable NUMA Memory information on a machine to determine available Memory resources before dispatching a job to a machine

We were noticing that jobs were being terminated by LSF with the message: TERM_MEMLIMIT: job killed after reaching LSF memory usage limit.Exited with exit code 137. Yet, this was happening before the job had even consumed the amount of memory rese...

over 1 year ago in Spectrum LSF / Scheduling 0 Future consideration

Please enter your email address

FILTER BY CATEGORY