Environment: We have Microsoft SQL Server 2016 (SP2-CU15) Standard Edition (64-bit) on Windows Server 2016 Standard 10.0. We have multiple SE instances on one physical server. We use processor affinity mask to tell which instance use what cores. Some cores are dedicated to instance, but most cores are shared between more than one instance. (Without using affinity mask we would not be able to use all physical machine power)
Problem: we have noticed that one instance is disturbing another one in case there is some heavy CPU load is running. Especially if that is one (!) big query running for long period of time. There always seems to be enough CPU power left available to affected instance not to experience visible slowdowns. I do believe that trace flag 8002 would help in this situation as described in this link.
What is the the wait stats (or other metrics) that I should monitor to ensure that trace flag really gives desired positive effect?