On Monday 27 August 2007, mcclnx mcc wrote:

We have several large projects will use ORACLE 10Gr2 database with CENTOS 4.X. Does anyone have performance report on 32 Bits O.S. (with 32 bit ORACLE) VS 64 bits?

Thanks.

Don't have anything recent on 10g and 4.x but we did some testing with a pre-release version of 11g running on CentOS 5... We started with 2GB ram on a Core 2 Duo E6400.

The results with our tests and data sets showed that the difference is nothing you will notice. In most cases the 64bit version was a tiny bit slower. However, two results were unusual. When processing outer left joins, the 64bit version was quite a bit faster - somewhere around 15%. The opposite was true on rollback of large transactions resulting from deadlocks. There the 32bit version was much subjectively faster. No exact timings though since I have no idea how to measure the exact time of a deadlock detection :)

Once you add more ram though or go to a larger smp system, the 64bit quickly leaves behind the 32bit version... All that said, we decided to go 64bit only for what we do here, even of small dev boxes. That makes the environment consistent and nothing we saw showed there was any real advantage to staying 32bit is your hardware supports 64.

Also, you will find other benchmarks online. There are some that show a large benefit of using 64bit (usually large memory systems) and others that show a benefit of staying 32bit (usually systems with small amounts of memory). Yet others show results similar to what we found. In the end it comes down to how much ram you have, your index types and some other things. If you need to know a good result for your scenario, you will have to do the testing yourself.

Peter.