On Thu, Feb 14, 2008 at 10:26 PM, Jim Elliott
<[EMAIL PROTECTED]> wrote:
> For the most part my view is from observation, not actual
> studies. And as Mark pointed out, we are talking about two REAL
> processors here with a Virtual 2-way running on it. IBM
> recommends that you always run Domino on a 2-way (or greater).
My apologies for over-editing my post while making the tone softer,
but it looks like the message still made it. Opinions and intuition
don't work well, especially with performance of Linux on z/VM where
very little is intuitive.
As David Kreuter points out, the thing with Domino is to compensate
for a design problem in Domino (and I believe I did play a role in
whatever started this hypothesis ;-)
> There are several places where you can get into a bottle-neck on
> a 1-way. I have also seen this at one customer with Oracle where
> the performance on a 2-way was dramatically better than on a
> 1-way (even though the total CPU utilization was about the same).
I am currently looking at data from a customer who is wasting 25% of
his CPU capacity because of these recommendations. When I discuss the
data with them, lowering the number of virtual CPUs will be one of the
first things to talk about.
> And from observation, I have seen this at other customers as
> well. With all the threads running in "modern" applications these
> days, using 2 or more processors seems to improve overall
> performance.
When you are with the customer after they installed their first two
servers, they may still have excessive resources. As long as they do,
more is better (almost always). On an idle system, tuning is trivial.
But when they grow their workload to what they bought the machine for,
things change dramatically.
The recommendation of "same number of virtual as real" is &bogus and
has confused people a lot. Imagine the customer running several dozen
virtual machines and wondering whether the poor performance was
because he forgot to give each of them 2 more virtual CPU's when they
got two extra IFLs...
The proper recommendation would be "certainly never more than what
you're likely to get when you ask for it" - which is obviously much
harder to answer without measurements. It depends on the number of
virtual machines and the workload. That's why I claim:
"When you don't know, one will do. When you have measured, probably too."
Refer to my other post earlier this week about the same subject.
Rob
--
Rob van der Heij
Velocity Software, Inc
http://velocitysoftware.com/
----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390
