In Test-4, as shown in the following figure, we increased the total number of SQL MIs running HammerDB TPC-C workloads to eight. These SQL MIs were the combined instances from Test-2 and Test-3.
We observed a flattening out of the TPM numbers across the different instances despite having separate resource configurations running. One of the best TPM numbers was observed from the four CPU requested instance, sqlmi-04, with 20 virtual users.
The first four SQL MIs had four vCPUs requested, and the second four SQL MIs had 6 vCPUs requested. With the combined workload profile, we saw greater efficiency and potential Azure consumption cost savings by potentially aligning with a four vCPU configuration for this workload pattern. With SQL Server, it is always best to maximize work aligned with an efficient core count.
For SQL MI specifications, see Test groups defined.