JPAB Benchmark
I've used jpab project.
Computer used is intel i7 PC (3770, 3.4 Ghz), Windows 8, 8 GB RAM
mysql version 5.6 (installed not tuned) hibernate: 4.3.5
JPAB runs few tests - I've selected one - were database index is used (to perform LIKE Query). (IndexTest from JPAB). (Other tests would be to unfair for mysql/hibernate).
JPAB for Airomem was adopted as little as possible(Airomem is not JPA provider so I could not simply copy airomem libs in it).
#Results table
There were three runs made with random operations (
5 threads- 5 operation batch,
5 threads- 5000 operation batch,
1 thread- 5 operation batch (for airomem batch size does not make difference (yet)).
Second and this column show operations per second. Comparison is AIROMEM/MYSQL
| Operation | MYSQL+HIBERNATE (op/s) | AIROMEM (op/s) | COMPARISON |
|---|---|---|---|
| PERSIST | 285.3280554 | 27526.11948 | 96.47182936 |
| RETRIEVE | 34995.8475 | 15,979,925.25 | 456.6234679 |
| QUERY | 6927.456138 | 2003843.444 | 289.26108 |
| UPDATE | 286.5544938 | 72729.09542 | 253.8054611 |
| REMOVE | 542.7794454 | 55774.73971 | 102.757649 |
| PERSIST | 4586.467796 | 28756.68658 | 6.269898288 |
| RETRIEVE | 359489.0272 | 0 | 0 |
| QUERY | 38.24162338 | 57023.14338 | 1491.127686 |
| UPDATE | 3103.522611 | 57023.14338 | 18.37368389 |
| REMOVE | 16710.30314 | 59188.26622 | 3.542022293 |
| PERSIST | 115.8041248 | 22415.60834 | 193.5648525 |
| RETRIEVE | 9163.032914 | 4020465.99 | 438.7702225 |
| QUERY | 1613.542369 | 1881612.139 | 1166.137422 |
| UPDATE | 112.5349315 | 45536.27554 | 404.6412516 |
| REMOVE | 107.7230397 | 2666.627712 | 24.75447889 |
Measurements show mostly that traditional benchmarks do not make a sense for airomem. Speed is between 3.5 times (worst case were there was 5k batch at mysql part) to 1400 times faster. But:
Initially I was getting strange results while measuring Airomem - because the system runs in loop every operation (like randomized query) for 30 seconds and counts successfull actions. Counter is int - wih airomem speed it meant int overflows :-). So I had also to limit maximum actions for airomem.
In one test operation count on airomem is 0 ... because it was so fast that in so called warmup time (see jpab code - int range was already exceeeded....).