Re: Read ahead

David Williams wrote: > > In article <6efd6m$nu0@bgtnsc03.worldnet.att.net>, Jay Aymond > <jay@ulife.com> writes > >Any recomendations on what percentage of buffers should be used for "ra_pages" > >and "ra_threshhold" in a system that is mostly OLTP and batch processing? > > > > > Not really, run onstat -p and total the first 3 RA columns at the > bottom, the total should be clsoe to RA-pages used. I'll just add two things. First you can increase the Read Ahead until the totals that David mentions no longer match then back down a little. Second, the values that work for you will depend quite a bit on the types of queries you do (if read cache % is high you can probably use Read Ahead to good advantage if low your queries are not likely to be sequential enough to gain from Read Ahead) and on the speed of your disk farm. If your disks are very fast or you use a large stripe set you may find that you only need a very small read ahead and a smaller read ahead threshhold and that increasing either will not only not gain you much but may increase I/O system calls unneccessarily. Conversely, if you use slower and/or single disks then read ahead is VERY important. We use 20-30 disk stripes of 7200RPM and faster FAST-WIDE SCSI 2 disks and get great performance from read ahead values of 16 and a threshhold of 4. Likewise at my previous job we used only a 4 drive stripe of 5400 RPM Narrow SCSI-2 drives and needed readahead of 64 and threshhold of 16 to keep up with a MUCH smaller load. Art S. Kagel