Skip to main content

In the link below I read:

"AHV lets you configure vCPUs (equivalent to CPU sockets) and cores per vCPU (equivalent to CPU cores) for each VM. We recommend first increasing the number of vCPUs when a VM needs more than one CPU, rather than increasing the cores per vCPU. For example, if a guest VM requires four CPUs, use four vCPUs with one core per vCPU."

From my own experience, performance tests, and NUMA recommendations, I know that you should usually increase cores per CPU first. When you reach the physical cores per socket, then add a second socket.

Can someone familiar with the deep architecture of AHV explain why we see that recommendation? I do not believe that an 8-vCPU VM configured as 2 sockets with 4 cores each will perform better than 1 socket with 8 cores.



https://portal.nutanix.com/page/documents/solutions/details?targetId=BP-2029-AHV:nutanix-ahv-cpu-configuration.html

the only thing that tends to determine what we do is licensing issues so for instance with SQL and support for sockets so 4 sockets with SQL std but support for 24 cores, etc.  So if you configure a SQL Std server with 6 sockets and say 2 cores per socket it will only use 4 sockets so you would need to deploy as 4 sockets with 3 cores instead.

the only thing that tends to determine what we do is licensing issues so for instance with SQL and support for sockets so 4 sockets with SQL std but support for 24 cores, etc.  So if you configure a SQL Std server with 6 sockets and say 2 cores per socket it will only use 4 sockets so you would need to deploy as 4 sockets with 3 cores instead.

Khoskin, I understand the point for specific use cases. My question is about the general workload.

general workloads I just follow the best practice which is to add sockets not cores

Terms & ConditionsAccessibility statement