Re: Veritas Striping for OLTP?

My personal opinion is that if you are truly using the DBMS in an OLTP environment, your performance is ENTIRELY a function of your memory buffer hit ratio. The only time I/O bandwidth is an issue is if you are doing s#itpiles of I/O, ie big batch jobs or DSS. Buy the memory (its cheap these days, dont skimp...), and configure the engine to keep the buffer hit ratio > 98% or so and life will be good. Greg p.s. Actually, if you guys have lots (> 50ish...) of small disks, some modest striping may be worthwhile. Otherwise, it's more bother than good, especially if you're mirroring, which in your business is probably prudent. You pretty much answered your own question in that table placement doesnt matter so much in OLTP environments, because memory buffer hits are the name of that game. Memory hits are orders of magnitude faster than any amount of striped I/O can ever approach. p.p.s. Michael - did you take an onbar class in DFW area fall of 1997?? Michael Benveniste wrote: > > Does anyone have experience with the use of Veritas striping on Informix > databases? We intend to use Informix in OLTP fashion, going after data > with indexes. We have few scan/join type queries. We would like to make > the lives of our DBA's simple and try to layout the data over striped > disks, separating data from indices. Beyond that we will just rely on > striping to give us a performance boost. > > My Informix SE says that striping is a bad idea, but his experience has > been on large DSS systems where explicit placement across different disks > and use of parallel queries helps performance. I'm not sure individual > placement of tables is necessarily that important in an OLTP application. > > Any ideas? Comments? Thanks and regards, > ---------- > Michael Benveniste > mike.benveniste@fmr.com > > Any comments or statements made are not necessarily those of Fidelity > Investments, its subsidiaries, or affiliates.