The following table shows the details of the scenarios evaluated using the TPC-DS benchmark tool where all 99 queries were run in random order:
Table 5. Test scenarios
VM | No of leaf nodes | vCPUs per leaf node | Memory per leaf node | Total vCPUs | Total memory |
Scenario -1 | 3 | 32 | 256 GB | 96 | 768 |
Scenario -2 | 4 | 32 | 256 GB | 128 | 1024 |
Scenario-3 | 5 | 16 | 160 GB | 80 | 800 |
Scenario-4 | 6 | 16 | 160 GB | 96 | 960 |
As mentioned in the Logical architecture section, there are four PowerFlex HCI nodes on which the SingleStore leaf node VMs were created.
The test was conducted starting with three leaf node virtual machines, 32 vCPUs per node, and 256 GB memory per leaf node. The number of leaf node VMs were increased gradually from 3 to 6 with a different number of vCPUs and memory per node to understand the variation in the performance. The execution time for the total 99 queries was captured for every scenario.
The current configuration supported the scaling of up to six leaf node VMs on four PowerFlex HCI nodes. Customers can add more PowerFlex nodes relative to the SingleStore leaf node VMs.
Note: SingleStore does not recommend overcommitting the vCPUs for the leaf nodes.