...

Conclusions

In all attempts, the Hearth FHIR server has better performance than the HAPI FHIR server. Creation and update actions take a lot of more time in the HAPI FHIR than in the Hearth. The search actions have better results but still, Hearth server is faster about 4 times. I noticed that the HAPI FHIR server caches results (not know for how long) for search requestsboth FHIR servers are very fast because all response times are lower than 200 milliseconds. The hearth is significantly faster for creation and update operations - all response times are lower or equals 20 milliseconds. For read and search operations, both FHIR servers are on par and generally, their times are sub-50ms for p90.

HAPI FHIR server requires more memory on the start but also during tests the memory usage is significantly higher over the Hearth server.

The backing data store characteristics (e.g. durability, consistency, memory, etc.) were not considered in tests.

Because performance tests and FHIR servers were on the same machine there are some notable limitations to this analysis:

• network latency was not accounted,
• variations in the host OS were not accounted for,
• test scenarios are a limited to a basic load test that should be refined when the real world is better understood

• read and search operations are likely highly impacted as the test, especially for HAPI FHIR server, is against Locations that are fully cached:

  Code Block

  hapi_1 | 2018-09-05 13:42:12.190 [http-nio-8080-exec-7] INFO c.u.f.j.s.SearchCoordinatorSvcImpl [SearchCoordinatorSvcImpl.java:151] Search entity marked as finished hapi_1 | 2018-09-05 13:42:12.217 [http-nio-8080-exec-5] INFO c.u.f.j.s.SearchCoordinatorSvcImpl [SearchCoordinatorSvcImpl.java:301] Reusing search f36b5bb4-d58c-4836-83fe-448225372c85 from cache hapi_1 | 2018-09-05 13:42:12.228 [http-nio-8080-exec-5] INFO c.u.f.j.s.SearchCoordinatorSvcImpl [SearchCoordinatorSvcImpl.java:151] Search entity marked as finished hapi_1 | 2018-09-05 13:42:12.245 [http-nio-8080-exec-8] INFO c.u.f.j.s.SearchCoordinatorSvcImpl [SearchCoordinatorSvcImpl.java:301] Reusing search f36b5bb4-d58c-4836-83fe-448225372c85 from cache hapi_1 | 2018-09-05 13:42:12.250 [http-nio-8080-exec-8] INFO c.u.f.j.s.SearchCoordinatorSvcImpl [SearchCoordinatorSvcImpl.java:151] Search entity marked as finished hapi_1 | 2018-09-05 13:42:12.265 [http-nio-8080-exec-1] INFO c.u.f.j.s.SearchCoordinatorSvcImpl [SearchCoordinatorSvcImpl.java:301] Reusing search f36b5bb4-d58c-4836-83fe-448225372c85 from cache hapi_1 | 2018-09-05 13:42:12.269 [http-nio-8080-exec-1] INFO c.u.f.j.s.SearchCoordinatorSvcImpl [SearchCoordinatorSvcImpl.java:151] Search entity marked as finished hapi_1 | 2018-09-05 13:42:12.288 [http-nio-8080-exec-4] INFO c.u.f.j.s.SearchCoordinatorSvcImpl [SearchCoordinatorSvcImpl.java:301] Reusing search f36b5bb4-d58c-4836-83fe-448225372c85 from cache

  
  

Overall these current performance metrics are within the expected needs of an FHIR server that is apart of OpenLMIS for interoperability.

Performance results

The following zip file contains all result data of all attempts. Those results are also available in the following repository.