Hi Salman,
in contrast to Stefan's reply I would recommend more redo switches per hour. I would try to achieve six to twelve switches per hour, this means every five to ten minutes. Normally the amount of logs written during a day is not constant. Therefore you should check how many logs your system produce during high workload (batch or dialog).
On average 15-20 minutes seems to be ok. You should also check how long your checkpoints are. Normally they should not be much longer than one or two minutes. As a first guess I would assume that Oracle produce the same amount of logs as Informix.
The number of groups is from my point of view dependent of your file system layout. If you have seperate disks for redologs (out of performance reasons) then it makes no sense to save the disk space, build as many groups as possible. There is absolutely no benefit keeping free space on the log volumes. The same is true if you run on a SAN box, you probably get much more than 400 MB in a single volume (in our environment 8GB is the smallest allocation unit).
The answer to you third question is complicated as Stefan already mentioned. If you ask performance enthusiasts, you'll end up in using raw devices. You save some CPU cycles because you run less layers from application to disk, but the question is: Is it worth it? My answer would be the same as Stefan's. I would go for cooked files. I think the CPUs today are fast enough to use cooked files instead of raw devices.
Kind regards
Ralph
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.