Score:0

Optimizing CPU and Memory use on Multi-socket NUMA server

ci flag

I am building a server that will be used by interns in my company to connect to and perform some tasks. The server is running AlmaLinux 9 and using KVM to run Windows 11 guests.

The server has 2 CPU sockets, each running 8-core CPU's.

After reading about this for hours, I wish to ask about topology.

What is the best topology for CPU's in the Windows 11 VMs?

On one hand I thought that since I can explore L3 cache advantages and so on by having all vCPU's run in one socket, I should choose only one socket in my topology and assign the necessary cores to it. On the the other hand, I thought that by giving multiple sockets to the VM's I am more closely mimicking the host system, which is supposed to be good, and scheduling of tasks from the guests may work better. However, in that case, does it not defeat the purpose of using NUMA on my host system since it introduces latencies depending on where cache is being accessed, for instance? Or does it make no difference at all since the VM will simply run as a process on the host?

I have read a zillion guides on these topics at this point and do not understand correctly still how they affect one another, so any help would be appreciated.

Alexander Pavluchenko avatar
pl flag
tools like numactl and lstopo can help you understand your current NUMA topology and plan your VM deployments accordingly.
djdomi avatar
za flag
mostly it depends on the scheduler but imho, remind that you can overshoot due you already assign at least 2 cores per vm. more interesting wozld be sharing the ram in case of ram pressure
I sit in a Tesla and translated this thread with Ai:

mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.