Hands on: Raid controllers

In recent Performance articles, we compared the features and performance of several popular Raid technologies using up to four hard disks. This time we’re following up those reports with tests using a more sophisticated hardware Raid controller.

Software vs hardware
Many motherboards and budget host adapter cards offer the choice of Raid Levels 0, 1 or 5. Raid 1 makes an identical copy of one disk on another to provide 100 per cent redundancy, albeit at the cost of half the total capacity and slower speed than Raid 0. Raid 5 uses three or more disks and delivers 100 per cent redundancy by writing parity data on all of them. The advantage over Raid 1 is only losing one disk worth of capacity to redundancy.

Raid 5 sounds ideal, but the calculation of parity data is complex and normally offloaded to your main processor by common motherboard controllers. It may be cheap to implement, but it results in greater overheads on your main CPU and often very poor write-speeds.

Hardware Raid controllers solve this problem by employing their own processors to take care of the necessary numbercrunching for Raid 5. Unsurprisingly, the additional components increase the cost of the controller, but that’s the price you pay for ultimate Raid performance.

We compared the Promise Fast Trak TX4310 tested in earlier editions against the Promise Supertrak EX8350. The TX4310 supports up to four serial ATA (Sata) hard disks and offers hardware Raid 0 and 1, but uses your main processor to deliver software-based Raid 5.

The EX8350 supports up to eight Sata hard disks and features a 500MHz Intel IOP333 XScale processor with 128MB of DDR memory to deliver accelerated hardware Raid 0, 1, 5 and even 6, which protects against failures on two disks at the same time. The EX8350 needs a PCI Express x4 slot, but will also work in the x16 slots used by graphics cards. The TX4310 and EX8350 cost £84 and £250 ex Vat, respectively, and are available from www.span.com.

For consistency with previous issues, we used the same test system, built around an Asus N4L-VM DH motherboard, with an Intel Core Duo T2600 and 2GB of Crucial Ballistix DDR2 memory. We used four identical Seagate 300GB 7200.9 Barracuda hard disks for each Raid array, but booted Windows from a separate Seagate 400GB 7200.8 disk.

We used a separate disk for the OS because the HD Tach benchmark can only perform write tests on unpartitioned volumes (www.simplisoftware.com). Each Raid array was built using the Promise adapter’s Bios controls, and we ran HD Tach’s Long Read mode with write testing enabled.