Performance Recommendations: SAS on Virtual Machines vs. Physical Servers

[arfajobs]

Hi Everyone !

Does installing SAS 9.4 on an on-premises virtual machine (VM) can indeed impact performance ?

What are the performance recommendations for running SAS on a virtual machine (VM), and how does it compare to installing SAS on a physical server in terms of efficiency, resource utilization, and scalability?

Best regards

Nidhal

[ChrisHemedinger]

Hi @arfajobs ,

The brief answer is Yes, installing SAS on a virtualized environment can affect performance. The SAS support policy promises functional compatibility within VMs but performance challenges may require additional planning and tuning.

This SAS Note links to several papers that go into lots of detail. Results will depend greatly on the resources you allocate to the VM and how you configure SAS to access those resources.

You don't mention "how much SAS" you're putting in the VM. Is it just a compute server, or are you including all of the metadata, midtier and data as well? All of these will affect the performance. You may be able to get the performance you need, but it may require some analysis and planning per your workloads.

[JuanS_OCS]

Hi @arfajobs ,

it is all indeed as @ChrisHemedinger explained.

Always when you add (virtualization) layers, you are placing an overhead of some degree: with virtual machines, with containers, cloud instances, etc.

The question is how much is significant and acceptable for your processes. For instance, all virtualization methods mentioned above are generally accepted nowadays - yet do mind that the penalty on performance still happens. The question is how much noticeable is, overall.

The only way to know is by having a good analysis of your workloads (SAS & non SAS) intended on those servers, and by aligning with an specialized representative from vendor or partner, whose most likely will ask you to expand on your information.

[SASKiwi]

Where I work all SAS 9.4 installations now run on VMs (VMware).  Even SAS Visual Analytics runs on VMs with an app server having 16 cores and 256GB of memory. It runs faster than the previous dedicated hardware as it uses the latest data centre server hardware. In my experience the VM overheads are a lot less significant than they used to be and can typically be ignored for sizing purposes. Of much greater importance is provisioning fast (high IO) data storage.