It seems that another alternative could exist, where the machine gets exclusive use of the pCPU and the HT core of that pCPU. The disadvantage to this is that it seems that the HT cores are either shared (as per default policy) or disabled (when using the None policy). The workload was running as a standard workload before, with a fixed amount of VMs on the host, low enough to enable them all dedicated CPU usage. The VMs are running a custom SIP/RTP workload, so latency sensitivitiy set to high is required for them to respond within the 20ms we need to keep SIP working correctly. Is there way to get a 2-core physical VM to get access to the HT cores and maintain the low-latency setting? Or is there an alternative way to obtain low-latency-like configuration without needing to use the low-latency:High setting?ĮDIT: More information regarding our setup. Now, if we add 2x more vCPU to 'attempt' to get 2 HT cores, the VM requires the CPU reservation to increase- which means it will grab another 2x pCPUs, instead of 2 vCPUs. Neither of those are the hyperthreading cores from the host.
When setting this VM up, it appears that it gets two cores in the guest. I'm having trouble putting together how to achieve the following with low-latency sensitivity virtual machines:Īllocate precisely required Mhz for CPU Reservation
0 kommentar(er)
