So what’s up this time. Our client has moved to Azure in classic lift and shift scenario. Well, almost. They’ve deployed new VM’s and installed SQL Server 2019 Standard in nice DTAP setting. The VM’s are standard E16-4as-v4 SKU. 4 vcpus and 128 GB memory. The disks are Premium SSD LRS ones with 2300 Max IOPS.
Their on-premises environment was a SQL Server 2016 standard edition running on a virtualisation layer with 128 GB of memory and 8 cores.
In both cases there are 2 numa nodes dividing the amount of cores between them.
Read on to learn more about the problem and what Reitse & co did to resolve it. Also check out the comments—Daniel Hutmacher, in particular, I think has the reason nailed.