Performance during Multithreaded Reading and Writing

We’ll run the threaded tests to check out the drives?ability to process multiple requests to sequentially located data. This emulates a situation when several large files are being read from the hard disk simultaneously.

To do this, we use IOMeter and time the test so that the amount of processed data was at least two times the size of the drive’s cache buffer. We create up to four simultaneously working threads (each thread is controlled by a worker with a separate disk space range). The disk is accessed in 64KB data blocks and the depth of the outgoing request queue is steadily changed from 1 to 8. The result is the combined read (write) speed of all the running threads.

The number of tables produced is huge, and you can view them here. If you don’t like to analyze them yourself, we can tell you that this test is won by Maxtor’s drives with a colossal advantage over their opponents.

The diagrams below show only the results achieved at a request queue depth of 1 as the closest to the typical load on the disk subsystem of a desktop computer.

Multithreaded Read

When processing one thread, the disks rank up according to their sequential read speeds. No wonder that the first two positions are occupied by Seagate’s 7200.10 drives. The 750GB model is on top since it is the “densest?drive in the series.

What’s curious, we also see Seagate’s drives at the bottom of the diagram (from the 7200.9 and NL35.2 series).

Let’s now see what we have when there are two read threads to be processed:

Well, what’s wrong with the Seagate drives? Can’t they switch to the second gear?

But Maxtor’s drives rise up to the occasion and show their very best. They have almost kept the same speed on switching from one thread to two. That’s what good look-ahead read algorithms can do!

The WD drives have slowed down, but not too much. The Hitachi drives have also become slower, and we can suspect that the smaller speed of the ATA-interfaced model is due to its smaller cache buffer (8MB against the SATA model’s 16MB) ?the difference is almost twofold.

Nothing changes with three or four threads. The Maxtors are ahead, the WD drives are doing their utmost to catch up with them, and the Seagate drives are working in random read mode.

Telling you the truth, we can predict the result of this test by ear, without even looking at the screen. If we hear the typical rattle of the read/write heads during this test, then we can be sure the drive won’t show a high speed. But if the drive is silent, its speed is going to be ok.

So, Seagate’s drives have problems here. They do not predict the load and their look-ahead reading is not optimal.