Re: Informix on RAID-5
Posted in 1996
It is also important to note that RAID-5 incurs more and lengthier recoveries than RAID-0+1. When a RAID-5 disk fails, it must be reconstructed based on parity, from other disks. This process itself is much slower than reading from a mirrored disk as in the RAID-0+1 case. For this reason, you are running 20-50% slower in RAID-5 normally, and when you have a disk go bad and a volume enters a degraded state, the system is slowed down additionally by this parity reconstruct issue. Also, note that as the RAID is reconstructed, in either case, and the system rebuilds the disk, the reconstruction of the RAID-5 disk will be slower than the reconstruction of the RAID-0+1 disk. In the latter case, it needs only to make an exact duplicate by consulting one other disk (or faster in the case of striped.) In the RAID-5 case, it has to consult other disks on a READ basis in order to reconstruct the failed disk after it has been replaced with a spare. If you run a 7x24 data gathering database, I do not suggest you go the route of RAID-5 at all. For me, RAID-5 would be good for just a standard NFS machine that serves a user community. I think for any mission-critical or 7x24 data gathernig operation (such as OLTP that runs round-the-clock), RAID-0+1 is a superior choice. By definition, mission-criticial has bottom-line fiscal impact and you should not be cheapening your support by going with RAID-5 which has substantial flaws that will impact any such mission-criticial/7x24 operation. Although RAID-5 appears to be less costly a choice at first glance at the beginning of a project as compared to RAID-0+1, most things that appear cheaper, end up being far more expensive in terms of repair and time lost due to repair as years of production pass... --Stuart Greg Moye <msmcl301.gmoye01@eds.com> wrote: >As cheap as disk space is, I think it's mostly a matter of personal taste. If you >need to write much, Raid 0 or 1 will be somewhat faster. If you mostly read, Raid 5 >provides a nice additional level of protection without consuming double the disk >space. Just make sure you have enough Raid5 dbspaces/chunks so that you can get lots >of page cleaners cranking in parallel for quick checkpoints. >Definitely don't Raid 5 the write intense stuff, that will really smart. >greg >Jeff Craig wrote: >> >> Does anyone have any sage advice regarding implementing an Informix >> database on a RAID-5 system? I am wondering how best to use Informix >> fragmentation in light of the fact that RAID-5 has already striped the >> data across disks. Since a RAID-5 "logical" drive is composed of several >> physical drives over which data is spread by the RAID software, should >> I try to fragment across logical drives? Should I skip Informix >> fragmentation for RAID-based data? I am not planning on putting the >> write-intensive stuff (logical and physical logs, temporary chunks, etc.) >> on RAID-5.snip.... >Greg Moye >Electronic Data Systems Corp. - Plano TX