Performance in Intel IOMeter DataBase Pattern
The Database pattern traditionally opens up the show. This pattern helps us check the drives’ performance with a stream of requests to read and write random-address 8KB data blocks. Changing the reads/writes ratio we can evaluate the quality of sorting the read and write requests.
Since the table is too large, we have to divide the results of this pattern into groups by the manufacturer.
For better readability, we draw graphs for different loads (the number of simultaneous read/write requests).
The two PATA devices behave in the same way, but the SATA model is quick to lose its ground. It definitely doesn’t like write requests. Curiously enough, the PATA and SATA drives from Hitachi are behaving absolutely differently in this test, quite contrary to what we saw in our tests of 160GB HDDs where drives with different interfaces, but with the same amount of cache memory, had the same speed in all the modes. Here, the SATA drive evidently has a less aggressive lazy write algorithm, and this is in fact easily explained: top-end SATA-interfaced products can be positioned as “hard disk drives for entry-level servers”, and aggressive lazy writing isn’t welcome for such an application.
The general picture has remained the same at request queue depth = 16. The SATA drive is much slower in all the modes where write requests are present.
Under the maximum load the SATA disk is just a little slower than the two ATA drives, which are still performing much alike to each other. The highest performance of the PATA models in the Random Write mode needs a separate mention: it is the first time drives with a spindle rotation speed of 7200rpm reach a speed of 200 processed requests per second.
The most curious fact, to our opinion, is that the SATA and PATA models from Hitachi behave differently, in spite of their having the same firmware version.