So we started looking for alternatives and we decided to try a RAID0+1 with IDE-harddisks just to see how well thatíd perform and whether the reliability was good enough for web server use. From our first set of performance tests we were simply blown away, a RAID0 array with two 80GB Western Digital 7200-rpm disks and a Promise FastTrak 100TX2 RAID controller
put our RAID0 array with two 10.000-rpm Quantum 10K2 disks to shame, and not by just a tiny margin. Throughput was almost doubled by adding the second disk, and latency was just a few tenths of a millisecond slower than a single disk, which was a world of difference from the SCSI RAID array that showed up to 25% degradation in performance.
Fig 2. The Western Digital WD1200JB harddisks mounted in the 2U-rack case hot swap brackets.
Naturally the IDE-RAID array had another big advantage over the SCSI-RAID array and that was the simple fact that heat wouldnít be a major issue, even in a much smaller rack-case. We initially planned on using a 4U-rack case with four SCSI harddisks, but the IDE harddisks would allow us to move to a 2U-rack case without any heat related problems. With help from Maxtor
and Western Digital
we acquired a total of four Maxtor D740X 80GB harddisks and four Western Digital WD1200JB 120GB harddisks for evaluation purposes.
We then proceeded with testing. Each RAID0+1 IDE-RAID array was run for two weeks on end, with close to 100% disk usage all the time. One array used the Maxtor harddisks whereas the other used the Western Digitals. We used a simulated rack-case environment where the disks were subjected to tests that were a close approximation of the database use in the real web server. After these tests it was obvious that IDE-RAID was a suitable solution for our web server. None of the eight disks had failed in two weeks of nearly 100% use, we had seen no heat related problems as none of the disks ever reached their maximum operating temperature. Nor did we experience any other problems with the disks or the array, it never faltered, it simply performed admirably.
Fig 3. The Western Digital WD1200JB harddisks seen from the inside.
From the results of our tests, it became quite obvious that the Western Digital 1200JBs
offered the best performance overall, and also ran a lot cooler than the Maxtor D740Xs. On top of that the Western Digital WD1200JB offered 40GB more capacity than the Maxtor D740X so our total RAID0+1 array would have a massive 240GB of storage capacity. That this RAID array is no slough either proved our second round of tests with a Seagate
Cheetah 15.000-rpm U160 SCSI disk, instead of the two 10.000-rpm Quantum 10K2 disks we tested with previously. And although it sounds ridiculous, the Cheetah was not able to keep pace with our RAID0+1 IDE-RAID array with four Western Digital WD1200JB harddisks. Clearly the 8MB cache on the Western Digital disks helped out a lot, as even in the latency, access time, benchmarks the disks performed admirably. The real difference however was the sustained throughput where the IDE-RAID0+1 array left the 15.000-rpm SCSI disk clearly behind.
So we obviously found a storage sub-system that was fit to suit our needs, although IDE-RAID wasnít something we initially planned on using as SCSI has always been the de-facto standard in these kind of applications, where performance and reliability need to go hand-in-hand. In hindsight, the new server has had over 150-days of uptime prior to the launch on October 1st, and we havenít had a single problem yet. We are happy that we didnít go the SCSI route; it saved us a lot of money and headaches. Four 10.000-rpm or 15.000-rpm harddisks and a SCSI RAID controller donít only cost a fair amount of money; they generate a substantial amount of heat which tends to drive up system temperature significantly in a small 2U-rack case.