Re: Diff between utlxplp.sql and utlxpls.sql ?

On Fri, 23 Dec 2005 21:07:03 +0100, Frank van Bortel wrote:

> I would like to see to rephrased to "Parallel execution may reduce
> response time" ; I have seen it actually increasing response time,
> even when processing 10M row tables, due to excessive waits on PX
> block sends. If that is a bug... I don't know.
> 10.1.0.4 on AIX 5L, 64 bit, 8 proc machine, 16GB, 4 CPU/8GB virtual
> partition in use.

Frank, I completely concur with your statement. PQO and parallel DML are meant for machines which have massive numbers of CPU resources. Today, operating systems are built with SMP in mind and the frontier at one time explored by then revolutionary VAX 782 is conquered. Result is that OS uses parallel CPU resources and it competes with Oracle RDBMS for them. Unless the are plenty of resources, results may be less then beneficial. I would recommend 16 CPU machines as a lower limit for seriously utilizing PQO. Also, the I/O throughput of the machine must be sufficient to enable many thousands of simultaneous I/O requests. That usually translates into FC/AL adapters, 32 or more gigabytes of non-volatile storage cache and RAID 1+0, BCV backups and alike. In other words, you cannot drive 40 tons of bricks in Ford F350. You need a bigger truck.

-- 
http://www.mgogala.com