During a VMware healthcheck at one of my customers I ran across the following error in /var/log/vmkwarning: “Memory is incorrectly balanced between the NUMA nodes of this system which will lead to poor performance. See /proc/vmware/NUMA/hardware for details on your current memory configuration.”
With the “more” command you can check how the memory is divided over the NUMA Nodes.
more /proc/vmware/NUMA/hardware
Output will look similar to this:
System type : Unspecific System
# NUMA Nodes : 2
Node ID MachineMem ManagedMem CPUs
0 01 8192 MB 7711 MB 0 1
1 02 4096 MB 4037 MB 2 3
The system was setup with 2 x 4GB and 2 x 2GB which results in this error. You can resolve this by changing the 2 x 2GB for 2 x 4GB, or you can disable NUMA in the BIOS. But I would prefer adding the correct size of memory. You would want to fix this because of the fact that when a processor needs more memory it will resort to non-local memory, in other words memory that was assigned to another processor. When using non-local memory data is transferred over the NUMA interconnect which is definitely slower than local memory.
More info on NUMA can be found on this wiki page. And there’s an article on VROOM about NUMA and performance.
The author uvazhuha! Here’s how to write blogs must! Everyone stands to learn, boys!