I do think that number of core per vCPU and vCPU number do not change anything about performances issues as, if configured identically, the same amount of resources are allocated to your VM. It was also confirmed to me by my instructor when I followed the vSphere course. Anyway, I will try to find a document about that point. I found this document on VMware's website : vSphere Documentation Center. It says that, from vSphere NUMA architectures may be exposed to VM. Using multiple vCore on 1 vCPU versus only multiple vCPU may increase performance if Guest OS and Guest applications are "NUMA-aware" and knows how to benefit from this. SQL must just know how to benefit from NUMA if you saw performance any differences.
Concerning your network configuration, you said 1 vSwitch and all 4 vmnics in it. You said failover ? Are all vmnic active at the same time or any vmnic standby (check vSwitch and all PG) ? What's the Load balancing policy ? Are you using Traffic shaping ?
You may check all your vmnic performance graph to see if one is overused compared to the three others.
Concerning %ready, it's a performance graph for your VMs. You will find it under the CPU section of Performance tab for each VM (also available for your vSphere Host). You should check this graph to see if the average value is too high. Post graph if possible.
Another question : how much memory do your 3 VMs have ?
Regards,
Ludo
Message was edited by: ldesfontaines
Added information about number of vCPU and number of vCores, all regarding NUMA architecture
If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful.
