IDE RAID Home Contact IFM

Overview

Storage size, speed (and reliability). When talking about drive space, the first two things are what people are most interested in. Anyone who has had a failed drive will also been keen on reliability (and so they should be!).

SCSI has been the traditional king for storage based arrays - and often the only option available. Recently I had a need for a lower cost solution that provided redundancy, in case a drive failed. I didn't want a software based solution, due to some horror stories I had heard in the past. So I started looking around at products available in NZ.

Solution

Two products came to light. The Adaptec 1200A, and the 2400A.

• Adaptec 1200A
  The 1200A is a "software assisted" IDE RAID card, supporting JBOD, striping and mirroring. It uses a hardware card to present a single drive to the system, so you can boot easily when a drive failure occurs, but uses a special software driver to perform the actual disk access once the OS has loaded (this card only works with the Windows based OSs). The card has dual ATA-100 channels and supports up to 4 hard drives.
• Adaptec 2400A
  The 2400A is a full hardware based IDE RAID caching controller (with 32MB on board by default, but only 16MB of this is used for caching). It supports JBOD, striping, mirroring and RAID5. It has four ATA-100 channels, and supports 4 ATA-100 hard drives. It is also an I2O based controller (causing it to be less CPU intensive). It also runs on a large number of OSs.

Now I had a dilemma. Should I go for the lower cost 1200A, or the more expensive 2400A. How much more performance would the 2400A deliver over the 1200A. So, I got both of them, and two Maxtor D740X 60GB 7200RPM ATA/133 drives (note that 7200RPM drives are about 50% faster than your standard 5400RPM drives, and only cost a little bit more - spend the extra, you wont regret it). I then constructed a mirrored and striped volume set using each controller, and measured the performance. I then created a load that simulated a single power user with heavy IO demands. Note that this test does not test the ability to handle concurrent I/Os, as would normally be experienced by a file server.

Results

When I first got the results, I had to repeat the tests several times, as I couldn't believe how good they were, for an IDE based system. As you can see below, the results are very comparable to entry level SCSI solutions - but at a fraction of the cost.

As can be seen, the 2400A delivered the best performance, and used substantially less CPU to do so. I put this down to the fact that it uses an I2O interface, and offloads a lot of the I/O processing to an onboard processor. The CPU utilization is important, because if you have a data processing application, you don't want all of the expensive CPU being tied up doing read and write operations to the disk subsystem. Also interesting to note was that striped I/O delivered 35% more throughput. I expected this figure to be higher, but put this down to the fact that I was not doing a concurrent I/O test.

The 1200A controller delivered performance equivalent to a single IDE drive in both its mirrored and striped configurations. The 1200A uses a software assisted approach, which was reflected in the far higher CPU load induced. I also believe this is why the performance didn't scale any higher - the 1200A is basically a normal IDE controller with a more sophisticated software driver to make the drive system more reliable, as opposed to achieving higher performance.

Recommendations

For low end applications, IDE RAID should definitely be considered. If you just want a more reliable drive subsystem, get the 1200A. If you want reliability, and a performance boost, get the 2400A.

Other points of interest are that the 2400A can support a hot swap IDE carrier. Also, 300GB IDE drives are available now. For less than $2500+GST, you could construct a RAID5 set using four of these drive to form an array just under a terabyte, with redundancy!