On Tue, Feb 24, 2009 at 2:59 PM, Alverson, Tom (Xetron) <tom.alver...@ngc.com> wrote: > ... mirror ... performance was fine ... > The RAID5 part was still horribly slow.
Short version: Your RAID controller stinks. Buy a better one. Explanation: Mirroring requires no computation; the controller just writes the same block to two different disks. RAID5 requires computation; an XOR operation has to be done for every byte of data in every block written to the logical RAID volume. If the RAID controller lacks fast, dedicated silicon to do the XOR calculations (i.e., it just uses a general-purpose microprocessor for XOR), RAID5 write performance will tend to be abysmal while mirroring is good. > some recommended adding the battery/more RAM option The battery will likely mean that the controller will turn on write-back caching, i.e., it will buffer pending writes in the controller's cache RAM, while signaling to the OS that the write operation completed. This is considered "safe" if you have a battery, since even in a server power failure or crash, the controller will keep the pending write around until the server is functional again. Doing this sort of caching can help write performance in some situations. In particular, a small burst of writes to random locations will perform much better. However, large/sustained writes will overwhelm the cache and kill performance again. -- Ben ~ Finally, powerful endpoint security that ISN'T a resource hog! ~ ~ <http://www.sunbeltsoftware.com/Business/VIPRE-Enterprise/> ~
