Re: Database and RAID

"Rob Prop" <rprop@concepts.nl> offerred: +This is a question that is commonly asked, and the answer is always the +same. The hardware vendors keep saying RAID is the fastest system. But it +isn't true. The fastest way is to use mirrored dbspaces (mirrored by +informix) because Informix writes in parrallel and for reading it can use +both disks the primary and the mirror. If the primary disk is busy the +mirror is used. A very common misconception. In fact, Informix has no way of knowing what disk is busy ahead of issuing the read, so it could not figure out which to read from. What is does is to logically "halve" the chunks. Reads from the first half are satisfied by the primary; reads from the second half are satisfied from the mirror. Problem with this strategy is that it assumes all of your chunk is full and that data is uniformly distributed across it. More commonly, on at least some of the chunks, the first half is more likely to be full than the second half, so the mirror is less likely to be employed for improving read performance. Generally, hardware mirroring is the fastest. It can usually take maximum advantage of optimized reads (as it can be handled at the lowest possible level). O/s level mirroring and Informix mirroring are about neck in neck performance-wise. There it is usually a question of who wants to handle the setup & maintenance, the Unix admin or the IDS Admin. Raid, in particular raid 5, is the cheapest solution as it requires less disk overhead (20%) than mirroring (100%). Dave Dave Kosenko davek@summitdata.com Director of Training Services (732) 469-4070 Summit Data Group (an Informix Authorized Education Center)