Sequential Read & Write Patterns

IOMeter is sending a stream of read and write requests with a request queue depth of 4. The size of the requested data block is changed each minute, so that we could see the dependence of the array’s sequential read/write speed on the size of the data block. This test is indicative of the highest speed the array can achieve.

You can click the following links for tabled results:

• IOMeter: Sequential Read results
• IOMeter: Sequential Write results

When reading from the RAID0 arrays, the three-platter models are all on the losing side. The Samsung is the best among them, the Hitachi being the worst one. The new two-platter drives are over 30% (and the leading Western Digital RE3 about 50%) faster than their three-platter opponents.

It is all clear with the losers which are all of the three-platter models. The results of the RE3 indicate that data is being read from both disks in a mirror, which leads to such an impressively high performance. Take note that this speed is achieved with 64KB data chunks, which is a very frequently used size.

It is all nice and tidy with RAID5: the three-platter drives are slower and go close to each other. The new HDDs enjoy a nice lead over them, the RE3 still being the fastest among them. Besides excellent firmware algorithms and fast read/write heads, the RE3 seems to have platters with somewhat higher recording density than the other HDDs have. We will check this out in WinBench’s data-transfer graphs.

It is the controller and its memory that affect the speed of writing in RAID0. The higher-density drives have but a small advantage. The Hitachi is the only obvious loser. The controller doesn’t work efficiently with it, and its top speed is much lower than that of any other HDD.

The HDDs split up into two groups when working in RAID10 arrays. Take note that the higher-density HDDs take the lead as soon as 4KB data chunks. It means that their advantage is going to be conspicuous in typical disk usage scenarios.

Writing is a traditionally difficult trial for RAID5 arrays. The combination of the controller’s and HDDs’ algorithms must be effective for the array to show good performance. The RE3 and the Barracuda 7200.11 are still in the lead, though. The Samsung and the ES.2 are good, too. The Hitachi and the Barracuda 7200.10 are slow.