09-22-2017 06:02 AM
In out current implementation, We are having server with RHEL having 256GB RAM.
Could you please suggest me values of below parameters in sasv9_local.cfg to utilize server resources at best.
Thank you in advance.
09-22-2017 08:32 AM
It comes down to how many concurrent SAS sessions you would have, and the approximate memory usage.
You want to avoid having the OS to swap, it's better if SAS does that for SAS sessions.
You can start with conservative numbers, monitor (OS, and SAS logs), and make adjustments as you go.
09-22-2017 08:43 AM
Hi Divyesh - What kind of configuration are you running? single? complex multi-teir?
I have a multi-teir configuration currently on RHEL 7.3 in AWS. I didn't set any local parameters in sasv9_local.cfg as I left the default settings at this time. I usually tell folks to create a sasv9.cfg file in their home directory that will override settings for analyses and processes that require more resources.
Hope this helps
09-22-2017 11:50 AM
Thank you for prompt response.
We are having plarform LSF implemented over cluster having one server node.
And it is two tier actually, database server and sas server are one different physical node.
09-22-2017 09:26 AM - edited 09-22-2017 09:32 AM
See this note for recommended values :
this isn't specific - as every case is (or ought to be) - so tuning up further for your context will take into account
SLA, performance requirements, user needs & expectations etc.
Usually, selecting the number of concurrent sessions and workload profile (light reporting, on demand analytics, batch etc.) as a whole or spliced up into 3-4 categories (light user/middle/heavy user + batch) will do.
With a large RAM size, if you don't expect too many concurrent sessions, then you might have a look at this paper which describes how to optimize RAM occupation for SAS jobs based on their priorities :
tmpfs virtual filesystem can help a lot.
If you have installed a SAS in-memory product like Visual Analytics then take care to confine each execution context to avoid conflict of resource like RAM...