Cores vs sockets

[adrian_vg]

Hi all,

Experimenting a bit with Nextcloud Server snap in a vm.

Noticed a significant decrease in wait states when going from four cores on one socket to ten cores on the one socket, as somewhat expected of course
Now, the node I run this vm on, is a dual cpu server.
I've noticed previously one can set how many sockets to use as well, in addition to how many cores, but haven't delved to much in the details to be honest.

So, to put it bluntly, is there any extra performance to be gained to use say five cores on two sockets, as opposed to ten cores on the one socket?

In what cases would it be preferred to use more sockets in a vm, instead of just setting how many cores you want to use?

How should one reason here?

Thanks for any pointers and advice in advance!

 

[leesteken]

If your hardware has non-uniform memory access (NUMA), then enable NUMA on each VM and use the same number of sockets as your hardware. This can/will improve memory performance/latency.

 

[adrian_vg]

If your hardware has non-uniform memory access (NUMA), then enable NUMA on each VM and use the same number of sockets as your hardware. This can/will improve memory performance/latency.

The servers are Dell PE's, a 710 and a 720.

How can I tell if they support NUMA?
I can't see from the tech specs on the Dell support site if they have it, unless I'm missing it completely.

Edit
Found this: https://vvirtual.wordpress.com/2011...ttings-in-dell-poweredge-r710-and-hp-dl380g6/
Will check this out.
Thanks @leesteken !

 

[LnxBil]

Found this: https://vvirtual.wordpress.com/2011...ttings-in-dell-poweredge-r710-and-hp-dl380g6/

Oh, I wasn't aware that you can configure this.

Generally you can check with numactl -s.

 

[adrian_vg]

Oh, I wasn't aware that you can configure this.

Generally you can check with numactl -s.

Cool, thanks for the command!

root@cyndane5:~# numactl -s
policy: default
preferred node: current
physcpubind: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
cpubind: 0 1
nodebind: 0 1
membind: 0 1

root@dragonborn2:~# numactl -s
policy: default
preferred node: current
physcpubind: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
cpubind: 0
nodebind: 0
membind: 0
root@dragonborn2:~#

Not sure how to read it though.
I'm guessing I need to set a particular policy maybe?
Or does the above indicate I'm good to go with enabling NUMA on each vm?

"Man numactl" is a bit lacking in explaining what policies are available and so on.

Update
On second thought, dragonborn2 above, the older of the Dells, and the 710 modell, seems to have only one CPU if I'm reading the output correct, or it needs setting NUMA to enabled in the BIOS.
Am I correct in thinking so?

 

[leesteken]

cyndane5 has two NUMA nodes (0 and 1) and dragonborn2 has one NUMA node (so no actual NUMA). You can always enable NUMA for VMs, it does not hurt. Use two virtual sockets for the first system and 1 virtual socket for the second (although it is also fine to use two here, if you want consistency or migration between them).

 

[adrian_vg]

cyndane5 has two NUMA nodes (0 and 1) and dragonborn2 has one NUMA node (so no actual NUMA). You can always enable NUMA for VMs, it does not hurt. Use two virtual sockets for the first system and 1 virtual socket for the second (although it is also fine to use two here, if you want consistency or migration between them).

Good info, thanks!