Real word tests: PostgreSQL database
I run the same Sysbench test on PostgreSQL also. Please remind that this article don't want to compare the two database softwares: its target is to show how the RAID layout affect their performances. So, lets start with the prepare test:
This time, RAID 5 is not much worse that RAID 10. The surprisingly thing, however, is to note how well the single disk setup copes with this task: it even outperform RAID 0. This is a completely unexpected thing and I need to conduct some additional tests to explain this behavior.
Now, its time for the one-million simple SQL query:
Now the standing are more or less in line with my expectations, with the exception of the RAID 5 results: while the simple Sysbench test is a read-only test, the results here are quite low. It can be related to the query planner's statistics update? Perhaps. Evidently, there is a reason why the PostgreSQL team explicitly discourage the use of RAID 5 with its database. Note that RAID 1 slightly outperform the single disk setup.
At last, we have the complex, read/write transactional test:
We have a similar picture here. RAID 5 is the slower, while RAID 0 and 10 are excellent performers. One more time, the RAID 1 setup is an hair above the single disk system.