RE: Require hard data on performance issues (correct email addr on this one)

Stay away from RAID 5. Writes are slow and some problems with the parity bit. Search this newsgroup for articles that Art wrote. Go with RAID 10. Spend the extra money. Well worth at the end. Make sure drives have write cache disabled !!!!. The raid controller with battery backup can handle write cache. Next issue on performance. First the more hard drives, the more spindles, better performance. 2 x 9 Gig drives better than 1 x 18 Gig drive with same RPM and interface. The next issue is how many drives to mirror and then stripe. The fastest drive out today {that I know of} is Seagate cheetah SCSI ULTRA160 {160Mbyte/sec} at 15 000 RPM, 3.9 avg. seek. The sustained transfer rate is about 40Mbyte/sec. {There is also an 2 Gb FC interface} So you take interface / sustained transfer rate = max number of drives to stripe. 160/40 = 4 dives. But then with a proper mirror where reads come from both primary and mirror, divide by 2 and test {writes perhaps better with 4 drives}. So in this case 4 drives, mirror 2 drives and then stripe. This will give you the max throughput for a single operation, but not necessary the max IO. That you will have to test. Also depending where you put your data. Spread IO across disks and fragment large tables where necessary. Next if you work with a disk pack like EMC, you will have to test yourself. Some of those packs have up to 4Gig cache and can handle big IO's. Hope it helps. -----Original Message----- From: Peter [SMTP:whitman@aottsd.uscourts.gov] Sent: Wednesday, May 24, 2000 6:44 PM To: informix-list@iiug.org Subject: Require hard data on performance issues (correct email addr on this one) The scenario is running Informix on a raid 5 verses a non raid system system. (Solaris 7 x86 OS) Could ~anyone~ point me to resources where there have been benchmarks run showing how much of a performance hit you take utilizing raid 5 compared to a non raid system?? And on a slightly different tack, are there any benchmarks showing performance gains using raw verses cooked file systems? -- again, reports rather then "it's quicker" TIA Peter W.