Re: Setting up Oracle DB to be a MTS.

Thomas Kyte <tkyte_at_us.oracle.com> writes:

> If you have 1,000 users and were using dedicated server and found that on
> average "select count(*) from v$session where status = 'ACTIVE'" returned
> 100 -- you would need about 10:1 to make it so no one has to wait for a
> shared server.

I don't understand this "no one has to wait for a shared server" goal. If the cpu is at 100% utilization what purpose is there to starting more queries simultaneously and having to context switch between them? The goal should be to reach 100% cpu utilization and then queue requests to handle them as efficiently as possible while keeping the cpu pegged.

If the wait times are too long and cpu is at 100% then you need more processors, not more shared servers competing for the limited cpu time. The question should really be therefore, how cpu intensive versus disk intensive your queries are. If your queries are mostly OLTP transactions hitting cache then I expect to need fewer servers per processor to keep the cpu utilization at 100%, I find 2-4 per processor is plenty, which is much less than the flat Oracle recommendation of 20 shared servers.

For disk intensive queries I suppose the 20 shared servers might make sense but I don't see the point of using shared servers for non-OLTP queries. I just have all the batch queries run in dedicated servers and tune MTS for pure OLTP. I wish there was a way to configure on a per-server basis parameters like sort_area_size and hash_area_size though.

--
greg Received on Thu Jan 13 2000 - 13:21:27 CST