RiskAnalytics: Performance Reference Case
A major goal for RiskAnalytics version 0.3 was to improve the performance.
In this blog I update the performance measurements which I first posted under the same subject in January 2009.
The performance test configuration was the same except, that we have a new dedicated server, namely a Quad Core AMD Opteron 1352 with 4GB RAM.
| Notebook |
Server |
||||||
| Iterations | 1 000 | 10 000 |
100 000 |
1 000 |
10 000 |
100 000 |
|
|---|---|---|---|---|---|---|---|
| One period |
11 | 76 | 776 | 11 | 54 | 479 | |
| Four periods |
57 | 509 | 9890 | 43 | 374 | 3690 | |
Conclusion:
- Simulation runtime performance is improved by factor 8 in the one period case and 3 in the four period case.
Remarks:
- We focused to improve the performance of one period examples. We will especially focus on multi-period models in a next release.
- Simulation time is linear. Our technology stack enables a high scalability.
- We added post simulation calculations for some statistical measures in order to speed result analysis. Actually post simulation calculations may take a while after a simulation has been completed.
- Our simulation engine uses currently only one core. We are currently investigating different technologies in order to use today's hardware capabilities. Even a recent notebook has a dual core processor. Hence, I am confident that there is still room for improvement.
Note that the runtime performance varies depending on other tasks running on the same machine. Unauthenticated users are currently allowed to run up to 10 000 iterations on the server.
-- Stefan Kunz
