If you just do the remount, and do NOT increase db_block_buffers, there is a better-than-average chance you'll have poor (worse) performance, and you'll end setting the mounts back to non-directio, thinking you need to look elsewhere for the problem.
Why? Because without directio, the UFS data is getting cached in memory, and your workload _may_ be benefitting from hitting in the file system page cache (on reads, and non-sync writes, although I'm not sure if Oracle does non-sync writes). By enabling directio across-the-board, you're turning off caching for all the file systems, which means all those memory page cache hits are going to turn into physical disk IOs, which, as you know, take about 1000 to 10000 times longer (i.e. reading a disk takes at least 1 millisecond - getting data from memory takes less than 1 microsecond). That is why it is necessary to tell Oracle to increase its memory cache, so the data that was in the kernel's page cache will get cached by Oracle. There are rare cases where the page cache was not a contributing factor to the performance of the workload, but that's less than 5% in my experience. /jim Hans-Peter wrote: > Ok thanks a lot sofar. > > We have planned down time for remounting the filesystems this weekend. > We will see what happens. > > Regards Hans-Peter > _______________________________________________ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
