Hi ,
Recently, we finished a POC project, one use case of this POC is to test the high concurrency performance of HANA.
The use case description is as follows:
we use java program to execute SQL by jdbc connection. The java program firstly start 5 threads to send request in parallel to 4 HANA node, and add 5 more threads at an interval of 5 seconds, the upper limits is 300 threads.
The result is very strange to explain:
In 4 node HANA cluster, we reached around 700 throughput/s, when we reduced the cluster to 2 node and 3 node the throughput decreased to 70 and 100 respectively. Does anybody have similar experience or can help to explain?
By the way, the hardware configuration is: 1T memory per node, 4CPUs and 15 cores per/CPU. The SQL executed on 2T raw data, which consumes 700G memory after loaded into HANA.
Thanks!