Raid uses multiple hard disks to deliver various combinations of greater reliability, performance or capacity.

We tested four configurations: Raid 0 with two disks, then three disks, followed by Raid 1 with two disks and Raid 5 with three disks. In each case the stripe block size was set to the default 16KB. Here we explain the performance results.

We used HD Tach to test each array’s performance. We ran the Long Read test with 32MB zones and enabled Write tests for the unpartitioned arrays.

For comparison we tested one of the 300GB Seagate disks on the motherboard’s non-Raid serial ATA (Sata) controller. This scored 54.5 and 46.4Mbytes/sec for average read and write speeds respectively, and 131.6Mbytes/sec in the burst test.

Raid 0 arrays have no fault tolerance, but what they lose in reliability they gain in capacity and performance. With two disks configured as a Raid 0 array, HD Tach measured the average read and write speeds as 100.5 and 72.9Mbytes/sec respectively.

The read performance was almost double that of the single disk and the write around 50 per cent higher, although the burst rate had fallen to 111.7Mbytes/sec.

A reconfigured three-disk array saw the average read and write times increase to 107.7 and 78Mbytes/sec respectively, and the burst rate remains about the same at 110.9Mbytes/sec. This was disappointing given that the addition of a third disk effectively decreases the array’s reliability.

We built a Raid 1 array using two disks. This delivered average read and write speeds of 56.1 and 43.8Mbytes/sec respectively, with a burst speed of 106.2Mbytes/sec. This measures roughly the same as a single disk.

Finally, we built a Raid 5 array using the minimum of three disks. Here the average read speed was a respectable 86.2Mbytes/sec thanks to striping, but HD Tach measured write performance as a paltry 7Mbytes/sec. Burst speed was 103.1Mbytes/sec.

Raid 5 arrays generally suffer from poorer write performance, but we felt perhaps HD Tach’s requirement of an unpartitioned volume for write tests may not be a fair measure.

To generate real-life figures we rebuilt each array, formatted them using NTFS, then timed how long it took to copy a 1.5GB video file from the boot disk. The Raid 0 arrays with two and three disks took 36 and 34 seconds respectively, while the Raid 1 array with two disks took 33 seconds.

After switching to the three-disk Raid 5 array, the same copying process took three minutes and 47 seconds, more than seven times longer than the Raid 0 and 1 configurations. Raid 5 arrays offer good read performance, but writing is slow.

In the Hardware column on Raid we detailed how long it took for the Raid 1 and 5 arrays to rebuild after a disk failure. We decided to retest their speed during the rebuilding process.

As partitioned volumes, HD Tach disabled its write tests. In terms of average read speeds though, the Raid 1 and Raid 5 arrays scored 8.4 and 18.4Mbytes/sec respectively.

Copying the same 1.5GB video file to each array took 47 seconds on the Raid 1 array and four minutes and 50 seconds on the Raid 5.

So both were slower in all respects, although, while the Raid 5 array was poor on writes, it still delivered a higher read score. Both arrays were more than capable of playing back Mpeg2 video recordings smoothly during the rebuild.