|
The SCSI protocol was created in the 1980's for use as a general purpose peripheral bus. It allows a variety of diverse components, such as tape drives, CD-ROMs, printers and disk drives, to share a single cable and communicate with the host adapter card using the same command language.
As usual in computer design, such high level compatibility comes at a major cost in performance. Even Ultra Wide SCSI has a theoretical limit of only 40 MB/Sec., which is drastically reduced when multiple disks are on the same cable. By comparison, the PCI Bus is rated at 132 MB/Sec., and RAIDZONETM has achieved over 80 MB/Sec. in real world benchmarks using ten UltraATA disks.
What has confused the performance issue for many years is that, up until recently, the fastest individual disks have been SCSI. This makes perfect economic sense, given that SCSI disks have always been a low volume, high priced product compared to ATA disks. Because they could get two or three times the price, disk manufacturers traditionally released their biggest and fastest drives in SCSI format first. This helped make SCSI disk arrays expensive, but couldn't overcome the fundamental SCSI bandwidth problem.
Building SCSI disk arrays is a no-brainer, and over thirty companies have entered and/or left that market in the past ten years. As disk performance increased, a single SCSI bus became clearly inadequate. This has led to complicated controllers with two or more SCSI channels, multiple cables, cache memory and elaborate on-card software, creating a single point of failure in a supposedly redundant system. |
|