Bruce, On 12/27/05 9:51 AM, "Bruce Momjian" <pgman@xxxxxxxxxxxxxxxx> wrote: > Historically, I have heard that RAID5 is only faster than RAID10 if > there are six or more drives. I think the real question here is "faster for what?" Also, just like the optimizer tunables for cpu/disk/memory speed relationships, the standing guidance for RAID has become outdated. Couple that with the predominance of really bad hardware RAID controllers and people not testing them or reporting their performance (HP, Adaptec, LSI, Dell) and we've got a mess. All we can really do is report success with various point solutions. RAID5 and RAID50 work fine for our customers who do OLAP type applications which are read-mostly. However, it only works well on good hardware and software, which at this time include the HW RAID controllers from 3Ware and reputedly Areca and SW using Linux SW RAID. I've heard that the external storage RAID controllers from EMC work well, and I'd suspect there are others, but none of the host-based SCSI HW RAID controllers I've tested work well on Linux. I say Linux, because I'm pretty sure that the HP smartarray controllers work well on Windows, but the Linux driver is so bad I'd say it doesn't work at all. WRT RAID10, it seems like throwing double the number of disks at the problems is something to be avoided if possible, though the random write performance may be important for OLTP. I think this assertion should be retested however in light of the increased speed of checksumming hardware and / or CPUs and faster, more effective drive electronics (write combining, write cache, etc). - Luke
