Re: Informix and RAID

Mickey Mestel wrote: > > craig, > > > Are there any whit-papers or other info on configuration of a data > > wharehouse and RAID? I will be configuring an 8.2 server and am > > considering hardware RAID over fragmentation. Any suggestions? > > i'm only going to touch on this, as it is a big subject, and will need > some research on your part if you aren't already familiar with this, but > informix fragmentation and the striping that happens with raid are not > the same thing at all. > > fragmentation takes a table and breaks it up into a number of smaller > tables, where you place each of those "fragments" into it's own > dbspace. the fragmentation has different strategies, but is usually > done on a column, (i.e. id > 0 and id < 100000 in dbspace1, > id >= 100000 and id < 2000000 in dbspace2, etc...). what this does is > allow the engine to search for the data and do any processing to the > data it recieves in parallal. if you have your table in 1 dbspace, the > engine will use 1 scan thread to get the data, if you have 157 dbspaces, > it will use 157 scan threads to get the data. imagine reading a > 100,000,000 row table from one disk, with one thread doing the work. What do you think is the maximum number of usefull scan threads if you have 4 disks ? Won't you agree with me if I'd say that 8 scan threads and 8 AIO-vp's would force the OS to perform disk positioning ? > now if you are using raid, then you have many disks doing the work, so > your data is coming back faster, but it is still being channeled through > 1 scan thread, 1 small pipe. if you set your raid scheme and your > fragmentation up correctly, you will have many disk heads bouncing > around ro return your data from disk, but it will be returning the data > to many threads that will actually process it. Let's say, if you have 4 disks available to store data on it, you should have 4 dbspaces. I/O balancing should be done by the RAID system, intelligent expression based fragmentation might be done by the DBA and should be detected by the Informix optimizer. > big topic, but you can't run a large data warehouse without > fragmentation. (of course you *can*, but you won't see nearly the > performance unless you do a lot of unnecessary stuff). > Bye Stefan Weideneder ----------------------------------------------------------------