All the test scenarios ran successfully, with good utilization of available compute, memory, and storage resources and some of the queries consumed 100 percent CPU utilization and 30 to 40 percentage of the memory.
The following figure displays the linearity when carrying out the tests.
The validation results indicated that the scenario-1 with three leaf nodes (32 vCPU and 256 GB memory) took 47 minutes and 3 seconds. In contrast, scenario-4 with six leaf nodes (16 vCPU and 160 GB memory) took 21 minutes 5 seconds to complete 99 TPCDS queries.
It is to be noted that the total vCPUs in both scenario-1 and scenario-4 were the same with scenaerio-4 showing better performance in terms of time that is taken to complete the 99 queries. This validation shows that having more leaf nodes with fewer processor cores has better performance outcomes.
Also, SingleStore database stores columnstore tables on disk, stores each column as a single unit, and stores segments of data for each column together in the same physical location. It was noticed that there is lot of disk activity happening during the TPC-DS test run and that the leaf node VMs with 16 vCPUs and 64 GB memory configuration show significant read bandwidth and IOPS.
Read Max BW
43.3 k IOPS
The following figure shows the PowerFlex read bandwidth per volume.