-
Bug
-
Resolution: Done
-
Critical
-
7.1.0.DR12
-
None
We noticed a significant drop in performance when running EJB client compared to EAP 7.0.
Short description of the testing scenario:
- many standalone EJB clients on three nodes spam a stateless bean (on fourth node) with requests (the number of threads on each node increases over time)
- the invoked method is just "echo", that means it returns the same value which was passed from client
- the value's size is 5 kB in "big data" scenario and 512 bytes in "small data" scenario
- the test ends when too many failures occur in clients (requests which fail with an exception)
- The perf test result is average duration of EJB calls
The numbers for big data scenario:
number of active clients | EAP 7.0 | EAP 7.1.0.DR12 |
---|---|---|
300 | 2 ms | 13 ms |
1500 | 10 ms | 100 ms |
3000 | 40 ms | 250-300 ms |
6000 | x | 300-500 ms |
maximum clients before test end | 4000 | 10000 |
Good news, EAP 7.1 can handle more active clients.
The numbers for small data scenario:
number of active clients | EAP 7.0 | EAP 7.1.0.DR12 |
---|---|---|
300 | 2 ms | 12 ms |
1500 | 3 ms | 90 ms |
3000 | 10 ms | 60-300 ms |
6000 | 120 ms | 200-500 ms |
maximum clients before test end | 10000 | 10000 |
What is notable is that with 7.0, the latencies increase quite steadily as the number of client increases, but with 7.1 they are very unpredictable, I saw cases where the latency reached 1000 ms with 2000 active clients and then dropped back to 200 ms after adding 1000 more clients...
- is related to
-
JBEAP-10520 Performance issue - remote JNDI lookups of JMS connection factory
- Closed