Hi,
Im not sure if i understand the CPU configuration on VM.
In Proxmox, you’re setting Sockets and Cores. This is Turenas and its “Experimental”. But I doesn’t understand the manual regarding this. Ref to manual, its seems you specify and dedicate cores. But that’s isn’t the plan.
Basically what i think i have done, but it not the result. With this setting, its shoukd gives me 80 cores total available for the VM. But that’s not the reality. Then i was think its was allocated cores, like VM should have 4 courses, and can use up to 40. But that’s doesn’t seems to be the result ether.
I have 2x Intel® Xeon® Gold 6140 Processor, should give me 36 cores and 72 threads.
I don t want to allocate 40 cores to the VM, i have only 38 and trueness need over half.
I just try to get the VM run smoothly, but its seems to max out on one core and not spread the workload on all cores.
On the VM, I can only count 17 coure.
HI, I’m not an expert here, but looking at the TrueNAS UI for CPU configuration states:
CPU Configuration
Enter the number of cores. Alternatively, use multiple values or ranges to set the CPU topology.
Example: 1-2, 5, 9-11.
In your screenshot, you have set CPU: 4-20, which is a range totaling 17, and that’s what shows in your VM.
If you want a specific number of cores in the VM, I think you should enter that number in the CPU configuration for your VM. It’s my understanding (I could be wrong) that in this case you can think of each thread as a core.
I don’t catch it, how 4-40 ends to 17 cores.
So you only setting core, and not threads.
How to get Maximum performers on the VM ?
Was on Proxmox before, and the same VM (ubuntu and exact same amount of docker container (same apps). Plex for example, run much smoother under Proxmox and feel it handle the gpu passthrough and transcoding better without the VM stop responding.
Now i manage some time to max out all 17 core at 100% …
The screenshot shows your VM is set to 4-20. This means that you have specifically told it to use CPU core #4 through core #20–17 total cores.
From the Instances tutorial you can either specify cores or just give it a number to work with
a. Enter the number of virtual CPU (vCPU) cores to allocate in CPU Configuration.
Set to an integer to expose that number of full vCPU cores to the instance.
Set to a range or comma-separated list to pin vCPUs to specific physical cores. For better cache locality and performance, select cores that share the same cache hierarchy or NUMA node. For example, to assign cores 0,1,2,5,9,10,11, enter 1-2,5,9-11.
So if you want it to use 40 cores, and you’re not particular about which, you can simply enter the number 40 in that field. Otherwise you can pick a range of 40 CPUs and specify.
Ah, okay
I misunderstand the range, fix and available free from the system. It’s a fix, but you can allocate specific core to the system…
Probably try to give it some additional cores and see if its run more smoothly.
You’ve got a pretty crazy number of cores. I had my box breathing easy with only 4, and you’re on this dual cpu setup with tons of cores and want it to run vm’s more smoothly. Can you be specific about your idea of smooth? Are you waiting on some processes, what could be better on your box?
You have an absolute wall of RAM and dozens of cores to allocate to your vm’s. Seems like it would run pretty fast.
Also consider that an amount of cores that exceeds one physical CPU will certainly lead to cache line flushes and potentially worse performance than with fewer cores that fit a single CPU.
Depending on the architecture a similar argument can be made for memory and PCIe lane access. You want to avoid having to use that inter CPU link.
I don’t know how well Linux handles NUMA architectures and how much control you get over it with KVM. Netflix did extensive work for FreeBSD …
