Hi,
I benchmarked performance of Infinispan embedded and HotRod servers by inserting and retrieving
Java objects. Here are the results:
Two Clients, Two HotRod Infinispan Servers, Two Nodes
Number of stored objects | Time to store 0.K objects | Time to store 1K objects | Time to store 10K objects |
---|---|---|---|
1000 | 2.226 | 2.438 | 2.797 |
10000 | 12.457 | 12.662 | 14.999 |
100000 | 54.778 | 57.963 | 73.021 |
1000000 | 466.090 | 495.431 | 624.543 |
Two Clients, Two Embedded Infinispan Servers, Two Nodes
Number of stored objects | Time to store 0.K objects | Time to store 1K objects | Time to store 10K objects |
---|---|---|---|
1000 | 1.403 | 1.332 | 1.543 |
10000 | 9.057 | 9.356 | 11.164 |
100000 | 46.284 | 47.412 | 57.649 |
1000000 | 390.677 | 411.569 | 531.239 |
System configuration:
1st server:
CPU – Intel Zeon E5405 2.0GHz
2 CPU with 4 cores each
6 GB main memory
OS – CentOS 5.5
JDK 1.6.0_33
Infinispan 5.1.6 FINAL
2nd server:
CPU – Intel i7-2760QM CPU @ 2.40GHz
1 CPU with 4 cores
2 GB main memory
OS – CentOS 5.5
JDK 1.6.0_33
Infinispan 5.1.6 FINAL
Network: 1 GigE
Do you think it is expected difference? Have you seen other results for this types of comparison?
Best regards.,
Jacob Nikom
Thanks for sharing this.
It looks about right to me - the in-vm access is faster for non-tx caches, but HR is quite fast as well, mainly due to its smart routing capabilities.