>> -----Original Message----- >> From: Steve Wechsler [mailto:smw@xxxxxxxxxxxxxxxxxxxxx] >> Sent: Thursday, December 11, 2003 9:53 PM >> >> With databases you want to maximize the number of spindles to >> increase throughput with minimal seeks (although with that many >> drives you're probably going to hit the limitations of the PCI >> bus). Many database servers will run an array with many smaller >> disks, rather than fewer large disks. It's much faster to pull 1 >> MB off 20 disks simultanously than to pull 20 MB off one disk. >> >> Also, someone commented on the negligible performance difference >> between SW & HW raid. While this may be the case for RAID 1, 0, >> or 10, with any RAID solution that requires parity computation >> (3, 5, 30 or 50), you had better go the hardware route. At the risk of starting a religious flame war, could you comment on the relative performance of hardware RAID in IDE vs. SCSI? For the sake of argument, assume top of the line hardware in both. Does hardware IDE RAID perform as well as hardware SCSI RAID, all else being equal (but is it ever?)? What factor does rotation speed play? That is, how will an array of 16 7,500 RPM drives compare with an array of 8 15,000 RPM drives? I assume the SCSI solution will be more expensive but how much? Will the performance make it worth the extra expense? There is a time to be cheap and a time to spend money, Proverbs 3:6 (more or less). The specific function for my array will be in a terminal server where many workstations are using the same drive array. The cost of the server is expected to be high but won't come close to the savings on workstation hardware. I can afford to be (cannot afford NOT to be) a bit extravagant on the server - fast CPUs, lots of RAM, big, fast RAID array... -- Henry Hartley
