Re: new RS/6000: from 8*250MHz to 2*750MHz - what to expect

It is possible that performance will actually get worse by doing this machine change.
I won't go into the arguments about why, I'll just point you to www.hotsos.com where Cary Millsap (he of OFA fame) has a paper that discusses why two batch jobs per CPU is the basic target. (This is relevant because parallel queries are basically batch jobs in disguise).

I can't remember the name of the article, but it will be fairly obvious when you get to the site. Read it, and then show it
to your manager so that he can make a
better informed cost/benefit decision.

--
Regards

Jonathan Lewis
http://www.jlcomp.demon.co.uk

Now available One-day tutorials:
  Cost Based Optimisation
  Trouble-shooting and Tuning
  Indexing Strategies

(see http://www.jlcomp.demon.co.uk/tutorial.html )

____UK_______April 8th
____UK_______April 22nd

____Denmark May 21-23rd

____USA_(FL)_May 2nd


Next dates for the 3-day seminar:
(see http://www.jlcomp.demon.co.uk/seminar.html )

____UK_(Manchester)_May
____USA_(CA, TX)_August


The Co-operative Oracle Users' FAQ
http://www.jlcomp.demon.co.uk/faq/ind_faq.html


"Erik Ykema" <Erik.SIHTEVOMER.Ykema_at_xs4all.nl> wrote in message
news:3e71263b$0$49099$e4fe514c_at_news.xs4all.nl...


> Esteemed Oracle practicioners,

> I would appreciate your thoughts on our RS/6000-change.

> Our operations manager has decided that our current RS/6000 (8 cpus

at 250


> MHz) machine should be replaced by a new one having just two 750 MHz

> processors. The cause is Oracle's license structure: per CPU.

> On this machine we have one instance for our datawarehouse (Oracle

8.0.6 -


> we have not planned to upgrade yet, AIX4): nightly batches in Pro*C

and


> SQL*Plus-scripts; during daytime it works for our OLTP users in

maximum of 7


> concurrent running reports.

> Lot of batch SQL has been tuned by trial and error (choosing the

degree on


> the tables; parallel hints) in order to have the batches fit in the

tight


> night processing window. Of course we will have to test everything

in


> pre-production phase. Additional tuning will end on my plate... and

I try to


> achieve some kind of work load planning for the coming half.

> From sar I have learned that, as usual, the disks are the main unix

> wait-events and that hardly ever, in batch or in OLTP, 100% CPU

utilization


> is reached. During batch this is also due to wait-constraints in the

batch


> job chain.

> What should I be aware of regarding performance? And any other

pleasant or


> unpleasant suprises? (I have roamed the c.d.o.* newsgroups but could

not


> find much insight.)

> Thanks,

> Erik Ykema

> Oracle Analyst/Programmer

>

>

>