This benchmark test was started with a single client running the test on 1500 warehouses. Then a second client was added, and the tests were run from both the clients each testing 1500 warehouses, with the total 3000 warehouses validated during the test. Finally, a third client was added, and the test was run from all the clients, each testing 1500 warehouses and the total warehouses validated during this test was 4500 warehouses. The tests were run by using the following commands and the results were captured accordingly:
Client-1:
# ./tpccbenchmark --nodes==<List_of_IP_address_of_database_nodes> --execute=true --warehouses=1500 --num-connections=300 --start-warehouse-id=1 --total-warehouses=3000 --warmup-time-secs=270 --initial-delay-secs=30
Client-2:
# ./tpccbenchmark --nodes=172.16.105.101,172.16.105.102,172.16.105.103 --execute=true --warehouses=1500 --num-connections=300 --start-warehouse-id=1501 --total-warehouses=3000 --warmup-time-secs=240 --initial-delay-secs=60
Client-3:
# ./tpccbenchmark --nodes=172.16.105.101,172.16.105.102,172.16.105.103 --execute=true --warehouses=1500 --num-connections=300 --start-warehouse-id=3001 --total-warehouses=4500 --warmup-time-secs=210 --initial-delay-secs=90
TPCC Metrics | PowerFlex Metrics | |||||||
Clients | Warehouse | TPM-C | Efficiency | Throughput (requests/sec) | Read Latency (ms) | Read IOPS | Write Latency (ms). | Write IOPS |
1 | 1,500 | 19,258 | 99.83 % | 713 | 0.30 | 5,790 | 0.42 | 1,484 |
2 | 3,000 | 38,495 | 99.78 % | 712 | 0.29 | 17,470 | 0.51 | 2,064 |
3 | 4,500 | 57,539 | 99.42 % | 706 | 0.29 | 28,480 | 0.57 | 2,457 |
It was observed that the average CPU utilization on each database node increased linearly with client scaling and reached 95% when the test was run from all the three clients. As the clients were scaled, TPM-C, PowerFlex read and write IOPS values increased linearly.
Note: The benchmark results depend on the workload type, hardware (VM or host) configuration, specific application requirements, and system design implementation. All performance data in this report was obtained in a rigorously controlled environment and there were no bottlenecks observed with the PowerFlex storage throughout the tests. Relative system performance results obtained in other operating environments may vary.