Database Patterns

In the Database pattern the drive is processing a stream of requests to read and write 8KB random-address data blocks. The ratio of read to write requests is changing from 0% to 100% with a step of 10% throughout the test while the request queue depth varies from 1 to 256.

You can click the following link to view the tabled results:

• IOMeter: Database, part 1
• IOMeter: Database, part 2

We will build diagrams for request queue depths of 1, 16 and 256.

The 3-platter Seagate 7200.11 really has no deferred writing at low loads. If its writing speed were even lower, we’d even suspect it to do write verification, which can produce the same effect as the lack of deferred writing. But we can see no double performance hit when we compare this drive’s writing and reading, so this is just a lack of deferred writing. A very odd behavior for a desktop product, we should say!

We can see the effect of the 32MB cache of the E7K1000 drive in comparison with the 16MB cache of the 7K1000.B. Its deferred writing is more efficient and it is almost as fast as the WD drives that take top places. Interestingly, among the WD drives the RE3 is somewhat better at writing than the Caviar Black although enterprise HDDs usually have worse deferred writing. Samsung’s team is an example: the F1 RAID version is slower than the desktop F1 TD at high percentages of writes.

Western Digital’s Green drives are not too bad, especially the 3-platter model with 32MB cache. The first model in the series, the 4-platter EACS, is the only one that is noticeably slower than the others. It is even slower than the similar RE2-GP model.

The Seagate SV35.3 is poor: it is not meant for server loads.

When the queue is increased to 16 requests, the HDDs can use request reordering algorithms. Western Digital’s pair is beyond competition still – they are going to be leaders of this review. The Hitachi E7K1000 is good but slower than the leaders. The 7K1000.B with 16MB buffer is far from brilliant: its NCQ is all right but it lacks memory for deferred writing.

Samsung’s drives show interesting behavior. The enterprise model has more efficient NCQ algorithms, but the desktop one has better deferred writing.

The 3-platter Seagate 7200.11 has a curious graph. It is comparable to most of its opponents at high percentages of reads, but slows down to 60 operations per second as soon as there are writes in the queue.

When the queue is longer, the 5400rpm drives find it difficult to compete with the 7200rpm models. The best of them (with 3 platters and 32MB of cache) cannot catch up with the 4-platter Barracuda.

The SV35.3 has poor performance again. It seems to have been deprived of both deferred writing and NCQ.

Western Digital’s drives deliver superb performance at the maximum load: 230 operations per second! By the way, it means that the drive performs two reads per each rotation of the platters. The Hitachi E7K1000, the best of the opponents, can only catch up with the leaders at pure writing.

Otherwise, there are no changes in the standings.

The huge load is good for the 5400rpm drives. Thanks to effective NCQ they get close to the 7200rpm models and look competitive to them. The WD10EADS with 32MB cache is especially good.