Home » Opinions » The Rules of RAID » RAID Performance - nForce 4 for Intel

RAID Performance - nForce 4 for Intel

This base run on a single drive using the integrated controller behaved much as we expected. The read and write speeds in HD Tach were slightly slower than the Raptor while the burst speed was slightly higher. However, as you can see from the results, the average read and write speeds will barely trouble an ATA66 interface, let alone ATA100, ATA133 or SATA150.

The file transfer test equated to a sustained figure of 30MB/second and the PCMark05 figures looked good too.

Next we added a second Seagate to the test PC and created a striped (RAID 0) array with a nominal capacity of 800GB and ran the tests again.

The HD Tach figures surged forward which is much as you would expect from a striped array. The PC is able to access two 8MB caches and as each cache is transferring data the other drive is working away in the background. Our file transfer test tells a different story, however, as the times barely changed. Compared to the single drive, the write time dropped by 11 seconds (seven per cent), to 2 minutes 18 seconds, while the read time rose by five seconds. In PCMark05, the HDD score rose by 28 per cent, while the overall PCMark05 score rose by just over two per cent.

Taken in conjunction with our file transfer tests the implication from the benchmark scores is that a striped array is a Good Thing but our real world tests show that in the real world you won’t see a massive increase in speed. There is a performance increase, but with just two discs it’s not nearly as great as you might expect.

When you think about this it may not come as a great surprise. A modern 7,200rpm drive has high areal density and probably has 8MB or even 16MB of cache, so although the basic design of a hard drive still looks like a jukebox it’s surprisingly fast for an electro-mechanical device.

Unless you really want to eek out every last bit of performance, you get very little benefit by running two drives in a striped array, particularly when you consider that the array has no redundancy and that each new drive adds an extra potential point of failure. If any of the drives or the controller fails then you’ve got a major problem, so at this stage RAID 0 looks a bit shaky.

In our next test we deleted the striped array and created a mirrored RAID 1 array. The theory says that mirroring is slower than a single drive as the controller’s workload is doubled but our tests showed that the performance drop-off is negligible and with RAID 1 you’re getting complete data protection for the cost of a second hard drive.

comments powered by Disqus