Impact of CPU Latency Sensitivity on all other VMs

[commguy23]

Out of a couple thousand VM's I have a few that are running latency sensitive applications and require the best performance I can provide. For these few VM's I have their memory reservation set to 100%, have the latency sensitivity set to high, and have a VM to host affinity rule keeping them on 1 host. It seems that DRS is keeping this host pretty vacant versus all other hosts (10 in this cluster) in the cluster after making these changes. There's still close to 1TB free RAM on this one host with these VM's set at 100% memory reservation, and about 10 other VM's running on the host. Most other hosts in this cluster are running over 100 VM's.

My question is on this; what impact does this setting have on other VM's, with that CPU latency sensitivity set, is it also reserving physical cores to vCPU's? I do not have CPU reservations set. It seems DRS balances CPU / RAM load across all other hosts accept this one, but I could not find documentation claiming physical cores are pinned to vCPUs.

[daphnissov]

Yes, when you enable latency sensitivity on a VM, it dedicates pCPUs to those vCPUs and changes the underlying scheduler. Because of this and other underlying changes, the density is going to be much lower on these hosts and that is normal. There are very few use cases for this latency sensitivity setting. The only ones I have seen are things like real-time trading applications and some NFV applications. Be very certain you have an appropriate use case and aren't enabling this setting simply because it sounds appealing.