Does unRAID really prefer to use core 0? - relevance for Threadripper 2

[testdasi]

It's sort of an accepted fact that unRAID prefers to use core 0 but is it really true?

Or is it that unRAID prefers to use the first available (i.e. not isolated) core?

 

In most system, first available core just happens to be core 0.

 

However, one can consider a scenario in which isolcpus is used to isolate core 0 for a VM.

Would unRAID skip to the first one that is not isolated?

 

I believe this seemingly rather academic question has implication for Threadripper 2 WX CPUs.

For these CPUs, 2 of the 4 dies don't have direct memory access - in effect creating 2 tiers of cores, the "fast" and the "slow", so to speak.

Given how things are typically numbered, core 0 is very likely to be a fast one, which is sort of a waste as typical unRAID tasks probably won't care too much about a bit more memory latency.

Hence it would not be a ridiculous idea to isolate all 16 fast cores for a "bang-bang" VM and let unRAID use the rest for more "mundane" stuff like Plex.

Would that even work? ?

 

TIA. ?

 

[John_M]

Surely in order to support your "bang-bang" VM you need QEMU to be running on a "fast" core too. What's the sense in relegating the hypervisor to a "slow" core?

[1812]

20 hours ago, testdasi said:

Would unRAID skip to the first one that is not isolated?

 

yes, it should.

[testdasi]

47 minutes ago, John_M said:

Surely in order to support your "bang-bang" VM you need QEMU to be running on a "fast" core too. What's the sense in relegating the hypervisor to a "slow" core?

 

From my (non-TR) server, when my VM is stressed 100%, the remaining core doesn't come anywhere close to that. I thought that means the QEMU doesn't really need to be that fast.

So I would assume any speed reduction by putting the QEMU on a slow core would be less than speed reduction by losing 1 fast core in the VM.

 

[John_M]

It's a question of latency, not speed.