Hi all,
I am making a few CPU benchmarks on an host equipped with a single 6c processor.
- CPU set with 1 core on 6 sockets : 5400 pts.
- CPU set with 6 cores on 1 sockets : 11800 pts.
Both setting were set to allow this VM to use an unlimited amount of CPU resources and with an high priority.
But what monitoring showed me, is that the multi-socket settings allow only a CPU usage of about 60% on the host while the multi-core use 100% of the host's CPU.
That explain the large result difference but why is that behavior ?
Thank you.
I rely on this article
Virtual Machine vCPU and vNUMA Rightsizing - Rules of Thumb - VMware VROOM! Blog - VMware Blogs
Basically in most cases use 1 socket and multiple cores
Even when thinking about "NUMA" these numbers doesnt make sense.
I would like to see some esxtop numbers.
Regards
Joerg
These numbers comes from PassMark CPU, of course they didn't means a lot.
But the double score by himself mean something, it mean the CPU as been fully used (verified by monitoring) and passed the math test a lot faster.
That's a fact, my question is why the first settings doesn't allow a full CPU usage ? ESX is probably managing a socket or a core differently ? I cannot find information about that.
I rely on this article
Virtual Machine vCPU and vNUMA Rightsizing - Rules of Thumb - VMware VROOM! Blog - VMware Blogs
Basically in most cases use 1 socket and multiple cores
Excellent article, it explain the behavior I had. It's probable due to the fact that PassMark make a better use of more core on a single processor than a lot of one core sockets.
VMWare should change the default VM settings when created because it's probable than most softwares work like this.
Thank you.