vmware NUMA nodes

[souzader]

Hi, 

my environment: 6.7 Update 3 running on an HPE BL460c Gen10

Both servers has 2 sockets with 24 cores per socket. In BIOS, sub-numa clustering is enabled and NUMA group size optimization is set to Clustered per best practice.

I would like to understand why in one server there are 4 NUMA nodes, this has been verified in ESXTOP as shown below, and on another server
there are just 2 NUMA nodes.

server 1

BL460c Gen10
2x Intel(R) Xeon(R) Platinum 8160 CPU @ 2.10GHz  24cores
8x DIMM(s) 32 GB. - 256 GB total memory

server 2

BL460c Gen10

2x Intel(R) Xeon(R) Platinum 8160 CPU @ 2.10GHz   24cores
4x DIMM(s) 64 GB. - 256 GB total memory

Is it because of the total physical memory installed?

[vmrale]

Hi,

it's interesting. I suspect that if there aren't any firmware version differences between servers and both of them have the same settings set in BIOS then it could be a specific of memory configuration for this processor and server model.

Check memory configuration guide for this server:

https://support.hpe.com/hpesc/public/docDisplay?docId=emr_na-a00027558en_us

Check the inner side of server housing cover there are usually some hints how to install memory. Maybe you will have to mix these DIMMs between these 2 servers to achive an equal NUMA node count you expect. You always can ask HPE Support Organization for help.

[depping · Accepted solution]

I suspect it has to do wit the memory configuration, or how the dimms are positioned. Frank wrote a very useful article on this topic:

https://frankdenneman.nl/2016/07/13/numa-deep-dive-4-local-memory-optimization/