Performance metrics such as TPM, IOPS, and CPU usage collected from HammerDB, PowerFlex Presentation Server and Vmware vSphere respectively are shown in the following charts.
The following figure showcases NOPM (new orders per minute) and TPM reported by HammerDB
Figure 5. TPM achieved with an increase in virtual users
The system achieved 2.01 Million IOPS when the benchmark was performed against 60 virtual users making queries on each database instance. The performance plateaued after this point and additional compute resources are required to continue to increase the scale.
Insights on how PowerFlex performed during tests can be gathered from the following figure:
Figure 6. IOPS and Latency observed in PowerFlex
Both IOPS and latency increases as the number of virtual users stepped up with each test. At 70 virtual users, PowerFlex reported a total IOPS of 262,000 with 0.39 millisecond read latency and 0.69 millisecond write latency. Additional compute only nodes would be able to support more virtual users. No bottlenecks were observed in the PowerFlex system during the tests.
The following figure shows the ESXi CPU usage:
Figure 7. CPU usage of one ESXi node with an increase in virtual users
It can be observed here that at 60 virtual users CPU usage reaches 85% and no significant improvement in performance was observed past this point. Comparing this behavior with other graphs in this section reveals that the CPU is the bottleneck in this system. Using a CPU with more cores or adding extra compute nodes to the system can increase performance.