database performance: Java Linux Databases: Multi-Threaded Combined
Multi-Threaded Combined Update/Insert/Delete/Select
This test starts an equal number of threads for each of the above operations.
Few threads
With less than 100 threads and no pause in each thread:
DB2, MySQL and Postgres all scale equally well.
Firebird starts well but does not scale as well as the first three with more than 50 threads.
SQL Anywhere is not too far behind Firebird, Informix aren't too far behind all of them, but consistently behind nonetheless.
Introducing a 1 millisecond delay in each thread allows for much higher initial throughput:
MySQL, Postgresql and DB2 perform equally well (unfortunately we lack data for DB2 with more than 200 threads).
SQL Anywhere is just as fast for up to 20 threads, then the performance starts to decrease, but it remains quite useable.
Informix does not scale well here, and the performance quickly drops by more than 60%.
Firebird was not tested.
Increasing the thread delay (not shown here) continued to increase the throughput but the rankings were not changed.
More threads
With hundreds of threads and a 5 millisecond pause in each thread:
MySQL is way ahead of the competition as it continues to scale even with thousands of concurrent threads,
Postgresql and DB2 are not far behind although the performance seems to peak at a few hundred threads rather than continue to scale up.
DB2 and Postgresql provide similar throughput, with DB2 edging past after 600 threads.
Firebird remains consistent with all the previous tests: providing reliable throughput.
SQL Anywhere is not far behind Firebird up to 400 threads and then continues its stable course - albeit 40% behind Firebird.
Again, Informix did not perform well at all in this test.
