The original Hardware Analysis web server, which has now been retired, was built from components that have proven themselves in our labs during months of testing in the normal process of evaluating computer hardware. This was a simple and foolproof way for us to make sure the components we picked were both reliable and rugged enough to withstand 24/7 operation. The new server, however, had to be designed from the ground up and needed to be at least twice as fast and more capable as the old one as well as offer reliability that is beyond questioning.

As most of Hardware Analysis content is database driven we had to make sure that we did not create a bottleneck there; databases require high-bandwidth throughout the system so we clearly had to design accordingly. Storage, memory and CPU(s) are all important factors here as a slow storage sub-system can slow down the fastest CPU, and similar for a high-speed storage sub-system that is slowed down by a under performing CPU. SCSI seemed the way to go for us as we’ve had excellent results with a 10.000-rpm Quantum U160 SCSI harddisk in our old web server, but we wanted RAID, RAID0+1 to be exact as that would offer the best combination between performance and reliability.



Fig 1. A peek inside the old Hardware Analysis web server featuring a 10.000-rpm U160 SCSI harddisk.

Naturally a RAID0+1 configuration requires four harddisks, which is rather costly if we want to continue using 10.000-rpm U160 SCSI harddisks and would also require a serious amount of cooling. And on top of that we’d need a dual-channel SCSI RAID controller that’s optimized for RAID0+1 to be able to truly benefit from the RAID array. From previous tests in our labs with a number of Adaptec and AMI SCSI RAID controllers we found that RAID0+1 wasn’t exactly their strongest point, we weren’t nearly getting the throughput you’d expect from a RAID0+1 SCSI array. There was too much overhead on the SCSI controller’s part and the latency, or access time, of the array wasn’t what we’d hoped for either.