Sure, i agree it is too general and that it could not even be an IO problem. my suggestion was to go raw for redologs since its good for redo write performance. Now, Direct IO is another beast. i do have my reservations about it but hey, its a free world. if Remigiusz can tell us that raw volumes for redologs aren't worth it since he tried it, then we have definitive proof. without knowing the disk array details, i really can't comment. i don't even know what application he has, what's the IO rate/pattern, especially redo rate. cheers anand On 22/02/07, Kevin Closson <kevinc@xxxxxxxxxxxxx> wrote:
What you really need is more spindles to do your writes. how you stripe them, blah blah is another huge discussion. try raw, will help. …that is too general a statement. Care to benchmark that? Direct I/O and RAW perform the same for sequential writes *most platforms). Cary and I just explained yesterday that it may not even be an I/O service time issue to start with.