VMware Cloud Community
rorybolt
Contributor
Contributor
Jump to solution

RESOLVED: Reduced capabilities going from v1.14 to v 1.15? - Not an issue

I have a 160 processor box with 512GB of memory... in v1.14 I was able to create VMs with up to 80 cores (the number of cores per processor). I could also assign more than 64 GB of memory to a virtual machine...

After upgrading to v1.15, these large VMs will no longer start unless I reduce the number of cores to 20 or less and the memory to 64 GB or less.

Are these limitations going to be relaxed again? Why the drastic reduction?

0 Kudos
1 Solution

Accepted Solutions
rorybolt
Contributor
Contributor
Jump to solution

Ugh! A BIOS update appears to have set the processors into quadrant mode instead of monolithic mode. This resulted in ESXi seeing 8 NUMA nodes (4 per socket) instead of 2 NUMA nodes (1 per socket).

It is a documented restriction that the Fling does not support core allocations or memory allocations spanning NUMA nodes. So previously, in monolithic mode I was limited to 256GB RAM (512GB total / 2 nodes) and 80 cores (160 total / 2 nodes)per VM, but in quadrant mode I was limited to 64 GB RAM (512GB total / 8 nodes) and 20 cores (160 total / 8 nodes). Changing the BIOS setting back verified this as the cause.

So sorry fo the false alarm - this had nothing to do with the 1.14 to 1.15 upgrade other than the coincidence of a BIOS update.

View solution in original post

2 Replies
cyprienlaplace
VMware Employee
VMware Employee
Jump to solution

Hi, thank you for reporting the issue, it was certainly not expected.

Can you share the vmware.log of the VMs that don't start?
Are they powering on and then failing to boot, or fail to power on?
0 Kudos
rorybolt
Contributor
Contributor
Jump to solution

Ugh! A BIOS update appears to have set the processors into quadrant mode instead of monolithic mode. This resulted in ESXi seeing 8 NUMA nodes (4 per socket) instead of 2 NUMA nodes (1 per socket).

It is a documented restriction that the Fling does not support core allocations or memory allocations spanning NUMA nodes. So previously, in monolithic mode I was limited to 256GB RAM (512GB total / 2 nodes) and 80 cores (160 total / 2 nodes)per VM, but in quadrant mode I was limited to 64 GB RAM (512GB total / 8 nodes) and 20 cores (160 total / 8 nodes). Changing the BIOS setting back verified this as the cause.

So sorry fo the false alarm - this had nothing to do with the 1.14 to 1.15 upgrade other than the coincidence of a BIOS update.