X-bit labs

Way to the Top: Hitachi Deskstar 7K1000 HDD Review (page 5)

Category: Storage

by Sergey Romanov

[ 07/12/2007 | 07:57 PM ]

Performance in Intel IOMeter

Multi-Threaded Tests

Besides the new testbed, we’ve got new tests. Making the hard disk read (or write) data sequentially into multiple address ranges, we check out the ability of its firmware to optimize performance by means of look-ahead reading and deferred writing.

The other HDDs’ diagrams can be viewed here .

Hitachi HDDs’ multi-threaded read speed is still only half their linear speed although every new model brings some improvement into this area. The Deskstar 7K500 is better than the others at processing two streams, but the newest 7K1000 is the overall leader.

It is a special thing with the driver of the Promise SATAII150 controller that the first thread blocks all the subsequent ones. When a large file is being read from the hard disk, the reading of any other file is almost impossible, which makes the system irresponsive to the user’s commands. This situation improves at a request queue depth of over 2 outstanding requests, but do you know many programs that ask for a second data chunk from the system before they receive the first one?

You can note that it is the typical behavior of HDDs without Command Queuing but the T7K500 is supposed to support that technology! So far we cannot see any optimizations of multiple requests in the two latest HDDs from Hitachi (equipped with the same processor version, by the way), but they are much better at reading three and four threads alternately.

So, Hitachi still has a lot of room for improvement. We do not check out the additional command set for controlling data thread, which was first introduced in the Deskstar 7K400, because we don’t know of an IDE controller that can utilize this feature. It means we have to wait for look-ahead reading algorithms to improve, especially as the Smooth Stream cannot ensure high speed on multiple data threads without them, either.

It is hardly better at writing: the first thread blocks the others at a request queue depth of 2 even on HDDs with enabled NCQ. This is surely our controller’s fault as we haven’t seen this thing happen on other controllers. Anyway, let’s see how performance of the Hitachi drives degenerates depending on the number of simultaneous write requests.

The HDDs have progressed. Each new generation copes with the threaded load better and better. The T7K250 and 7K500, the first HDDs from Hitachi to be equipped with Native Command Queuing, behave characteristically when processing 2 simultaneous threads. For the more recent models the performance hit is almost proportional to the number of threads and the Deskstar 7K1000 leaves no chance to its precursors irrespective of the number of simultaneous requests thanks to its highest linear speed.