Hey All!
I have an R730XD with twin E5-2690 v4 CPU’s.
I run VM’s in an NVME Z1 pool with 4 2TB Samsung 990 Pro’s.
I have 768 GB of 2400Mhz ram that I’m using for Cache as well as VM’s with a 99.99% hit rate.
VM’s Include:
HAOS
Database (large database functions)
System monitoring
MQTT with some RTL stuff running on it
LLM/Video/Motion/Plex… anything GPU intensive
I have some other tertitary VM’s that aren’t really doing much that aren’t important.
When monitoring the iostat for the pool the SLOG device (at 1 second intervals) the max hit I saw on the device was 14 MB’s
I verified that the current link speed matched the maximum supported of the x8 PCIe Gen3 of 8GT/s.
I do see that the SLOG is doing alot of very small writes often so at minimum it’s saving usage of the disks but it’s much slower than it should be.
I checked with sync writes always and standard and on large database hits the allocation never went abobve 8 MB’s (should have been much higher) and there was no appreciable speed difference between standard and always.
I confirmed which NUMA pinning the RMS-200 was on and also know that I am not pinning my VM’s so they are free to float. Should I be pinning my VM’s to specific cpus associated with the NUMA channel? If yes, should I be prioritizing my HAOS/Database and let my GPU intensive tasks that are AI, motion, recognition and so forth be free to float or should I pin those as well?