Hi Hans-Peter, There is an Oracle bug that is not fixed until 10.2.0.4 (if I remember right) where filesystemio_options is broken and so it won't do the correct thing with directio and ufs. The locks you are seeing are classic single-writer-lock issues and indicate that you aren't getting directio. The Oracle bug number is Oracle Bug ID 5752399 (Oracle setall w/ UFS).
You might check to be sure you have the fix to that. The only workaround is to use "forcedirectio" in the mount options, which has been pointed out has other issues for some activities, so getting the bug fixed would be a better idea. Regards, Dave Miller Hans-Peter wrote, On 12/04/08 06:55: >> Do you have directio enabled on UFS? >> Especially for the redo logs? > > That is the strange thing. > filesystemio_options has been set to setall which is asynch and directio. > But when I dtrace the fbt calls I see only directio calls for the database > writer > but not for the log writer. > > Perhaps we should mount the filesystems with forcedirectio. > Would noatime also help? > > Regards > >> With directio enabled, UFS writes to the log do not >> serialize on the RW lock for the log file(s). >> >> directio will also bypass the memory cache, so you >> need >> to increase the Oracle db_block_buffers when enabling >> UFS directio. >> >> Thanks, >> /jim >> >> >> Hans-Peter wrote: >>> Hi Mauro, >>> >>> Yes I understand why sysinfo is not the best to >> measure IO. >>> But I just wanted to see when in the whole trace >> the actual physical write was being done. >>> So it seems to me that, because the >> sysinfo:::pwrite is right at the end the performance >> bottle neck is more because of the locking etc. >>> The database files are on ufs filesystems. >>> Is zfs better? >>> >>> Regards Hans-Peter >>> >> _______________________________________________ >> dtrace-discuss mailing list >> dtrace-discuss@opensolaris.org _______________________________________________ dtrace-discuss mailing list dtrace-discuss@opensolaris.org
