SAS Grid - Define queues

[MariaD]

Hi folks,

We are restructuring our SAS Grid environment and we'd like to validate if the following scenario is possible to configure:

- A queue only for batch process;

- A queue for short duration jobs: any user with short duration jobs could execute on this queue;

- A queue for high priority users: only users in certain group could use this queue;

- A queue for high usage users (users with a hugh number of processo executed): only user in certain group could use this queue (Is it possible to limit the maximum usage for this kind of queue?)

- A queue for night executions: only users in certain group could user this queue to run process at night

- A queue for normal execution

Is it possible to configure this kind of scenario? We only want to know if SAS allows this type of configuration.

Additionally, we have multiples SASApp configured (different works areas for different group of users). Is it possible to have multiple SASApp with the suggested queue scenario?

Regards,

[Anand_V]

Hi @MariaD

I think most of these queues like short, night jobs etc are available out of the box.
You can restrict users on which queue they can submit their jobs.
You can configure multiple application contexts (resource in LSF terms). These are usually tied to hosts and then you can tie a host to a queue.

I think it’s possible to achieve most of these configuration.

[MariaD]

Hi @Anand_V ,

Thanks for your comment. One additional question. In this scenario, it is possible the same user uses multiple queues depending the case?

For example, in some situation the user uses short queues because his process allow to it. In other case, the user user night queues.. That mean, the user is able to automatically change between queues depending some conditions? 

[Anand_V]

Yes, it’s possible. Since you haven’t mentioned how the users would be submitting jobs via EG or Command line, depending on that settings would vary.

If the requirement is submit jobs to specific queues by default via EG then you may have to set up rules in SMC.

If the requirement is via command line then user can choose their queue while submitting job using bsub. Some examples here - https://www.ibm.com/support/knowledgecenter/SSWRJV_10.1.0/lsf_users_guide/job_submit_to_queue.html

[MariaD]

Thanks again @Anand_V . You are totally right, I forget to mention about how the user executes their jobs. Mostly its by SAS EG.

Is there any documentation/samples to how to set up the rules in SMC?

Regards,

[Anand_V]

This paper by @AndrewHowell is a good read to set the options. http://support.sas.com/resources/papers/proceedings16/11100-2016.pdf

[MariaD]

Thanks @Anand_V . One last question, is it possible to set a CPU limit (maximum % of utilisation) for a queue?

[Anand_V · Accepted solution]

Yes, there is a parameter for queues called CPULIMIT which can Limit the total CPU time the job can use. This parameter is useful for preventing runaway jobs or jobs that use up too many resources.

Check this link in CPULIMIT section: https://www.ibm.com/support/knowledgecenter/SSWRJV_10.1.0/lsf_config_ref/lsb.queues.5.html

[MariaD]

Hi @Anand_V 

One addition question. Suppose that we have 1000 slots to distribute between queues. We don't want to implement process limitation per user. 

If we create a queue for night executions with 200 slots. What's happen with this slots during the rest of the time? We can use this 200 slots in other queue? The other queue need to have time windows defined as well?

Or if a define a normal queue with all the slots (1000) and a night queue with 200 slots with higher priority. That's means when night queue start, the normal queue is automatically reduced to 800 slots? Or we need to define the normal queue without limit slots so the normal hours uses 1000 slots but when night queue start, the normal queue was automatically reduced to 800 (1000 less 200 slots from night). 

Regards, 

[Anand_V]

LSF re-balances the slot usage as and when the queue has less work or more work. Its not like that the resources will be idle for the 200 Jobs always. This link shares some good examples on how it does this - https://www.bsc.es/support/LSF/9.1.2/lsf_admin/index.htm?slot_allocation_examples.html~main

However if you define MAX no of jobs that a queue can execute that will be a hard limit on the queue and it will go to closed state if the NJobs running matches the MAX.