Re: Bloody Oracle support!!!

NetComrade wrote:
> .......
> We use Oracle 8.1.7.4 and 9.2.0.6/7 on Solaris 2.7 and RH4
> remove NSPAM to email

For those using RHEL3 and 9.2.0.6, here are some tips for really fast IO times:
1- Read about hugetlb_pool and setup for RHEL in note: 262004.1. Forget the very large SGA thing unless you have in excess of 32GB memory. Just go for hugetlb_pool, locked SGA and shmmax. Reboot to make sure you really got the lot. 2- Use directIO if at all possible after all the patching rigmarole. Must set large buffer size (watching the 32-bit limits) and set a suitable number of dbwrs - I ended up with 4. 3- Set the file systems to 4K block size with redo ones at 1024, make sure you mount all with "noatime" and "data=writeback". 4- Check the elvtune settings, something around -r 2048 -w 8192 is a good start, you may have to tweak this and use iostat to test the best balance. The redo log f/s may need tweaking with lower values. Go here for the sordid details if you have never played with this thing:
http://strasbourg.linuxfr.org/jl3/features-2.3-1.html

www.puschitz.com is also your friend but make sure you READ what's in there, don't just eyeball it! And follow up on the links and all doco.

And for Chrissakes:
*TEST*
before you make changes in production!

The darn thing ends up *fast*, particularly with a good cached RAID server on the other side. 80MB/s sustained *write* speed with burts up to 150 is not to be smirked at in this price range hardware. Sun, HP or IBM can of course beat that but at a much higher price point. Enjoy the ride, folks.

If any1 has any tips/hints on Oracle Net settings for 9ir2, I'm all ears: don't have time to review everything and I could use some speed-up on the comms side. Not connect times or number of connections, just raw speed for very large numbers of very large rows... Received on Mon Nov 21 2005 - 18:49:25 CST