Re: Concurrent IO on AIX

my 2 0.01 i would measure it. my experience is that raw is always faster and of course you have kaio. cooked uses unix filesystem cache and if i recall correctly thin ice....one can not set resident on aix??? since it uses fs cache you cache it more then once; due to fs cache you beat up unix!!! i did a compare on aix once using informix and char raw devs while another guy was doing JFS using dd. informix wrote at 50 MB/sec (10 disks/dbspaces 10 table fragments) JSF could write max 40MB/sec (striped across 10 disks) Superboer. Jonathan Leffler schreef: > Jürgen Buck wrote: > > I heard concurrent IO (cio) is as fast than using raw devices for DB2. > > Is this also true for IDS ? > > > > Is there any negative impact by using CIO instead of raw devices ? > > There's a document at > http://www-03.ibm.com/servers/aix/whitepapers/db_perf_aix.pdf about the > performance of CIO. > > Concurrent IO is an option to JFS2 - journalled file system. > Inherently, therefore, it is for use with 'cooked files' rather than > raw. Normally, you would not use a journalled file system for a > database - the journalling system provides roughly the same > functionality that a DBMS provides in terms of atomic changes with > recoverability, so you are overlapping the functionality. That said, > there's no reason you can't use a JFS for database storage. > > If your dbspaces are in cooked files on a JFS2 file system, I don't see > any particular reason not to use CIO. I've not validated that with > anyone - and I've not read every word of the paper, either.