This portal is to open public enhancement requests against products and services offered by the IBM Data & AI organization. To view all of your ideas submitted to IBM, create and manage groups of Ideas, or create an idea explicitly set to be either visible by all (public) or visible only to you and IBM (private), use the IBM Unified Ideas Portal (https://ideas.ibm.com).
Shape the future of IBM!
We invite you to shape the future of IBM, including product roadmaps, by submitting ideas that matter to you the most. Here's how it works:
Search existing ideas
Start by searching and reviewing ideas and requests to enhance a product or service. Take a look at ideas others have posted, and add a comment, vote, or subscribe to updateson them if they matter to you. If you can't find what you are looking for,
Post your ideas
Post ideas and requests to enhance a product or service. Take a look at ideas others have posted and upvote them if they matter to you,
Post an idea
Upvote ideas that matter most to you
Get feedback from the IBM team to refine your idea
Specific links you will want to bookmark for future use
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. The GSLA attempts to keep the usage of these balanced according to the ALLOCATION. However, it does not have a mechanism to force rapid re-balancing when one groups Demand surges, and other groups are using the starved group's share. There is a SIMPLIFIED_GUARANTEE, which supports PREEMPTION from one queue to jobs in another queue. But, the jobs from the groups will be sharing the same queue. Thus, the preemption needs to be between jobs in the same queue. Therefore, like with License Scheduler, we would like a ServiceClass to have an OWN, and be able to preempt borrowing jobs when the OWN'er needs the slots back.
Suggested Implementation: Each ServiceClass will have an OWN of a Pool. Not sure if that should be defined in the Pool Resources or the ServiceClasses. To define preemption paths: The mechanism that tracks the jobs (pool resources) used in each ServiceClass, will (for some ServiceClasses) have a string that says whether these jobs are PREEMPTABLE by (Sc1,Sc2 ..), or PREEMPTIVE to (Sc1, Sc2). And will have another string that defines the conditions under which preemption happens. For example, PREEMPT_FOR = "PEND 300", would mean that jobs in the ServiceClass that have been in PEND for 300 seconds BECAUSE of a lack of the pooled resources, will preempt preemptable jobs in other service classes that are over their share limit.
That would be the basics of the intra-GSLA preemption. There are, of course, a lot of caveats to consider. For example, jobs that are PEND'ing need to have a sole and only PEND reason being the lack of the pool resources. If they are also pending for Licenses, there is no benefit in them preempting jobs that already have licenses and slots. Assuming that the PEND'ing job's License resources can be met, then it needs to choose wisely the jobs to PREEMPT. That would be the jobs which, first order, were the last to use the excess of shared resources. That is, if the borrowing ServiceClass jobs have exceeded their share by 20 slots, then the jobs on top of that would be the first to be checked. The Second order check is to verify that the resources that the target job is using (the machine, memory) will satisfy the preempting job's needs. If not, there is again, no benefit to preempting it. Finally (third order check), concerns the Priority stacks of the candidate jobs. It is presumed that the job triggering the PREEMPTION is already the highest priority, since LSF chose it. But, the job to be preempted should be the lowest priority in the set of jobs that could be preempted. So, although the First Order check is to find the job(s) which have piled on top of the over-fed stack, the Third Order check is to sort them by their Priorities. Those priorities, I believe, are calculated from a complex formula involving the various priority ranking systems (Queue Priority, Fairshare, Job-Level, License Scheduler).
Do not place IBM confidential, company confidential, or personal information into any field.