On Tuesday, 02/21/2006 at 04:32 CET, Eginhard Jaeger 
<[EMAIL PROTECTED]> wrote:
> Note, however, that statistically relevant wait state information is 
best
> obtained from CP monitor data. And the easiest way to convert CP monitor
> data to numbers you can understand is to use a performance monitor ..

Trying to manage the performance characteristics of a z/VM system without 
using some sort of performance management kit is like entering the boxing 
ring with *both* hands tied behind your back.

CP INDICATE commands provide a snapshot (sometimes instantaneous, 
sometimes smoothed) of CP control block contents, really.  They don't give 
tell you how many times some Very Interesting event happened in some 
relevant interval of time and you don't see trends.  You can see "I have a 
problem", but you can't readily identify its source.  Granted, some people 
with years of VM experience can ferret out the source, but its a rare 
gift.

It's the same reason I still teach TCP/IP routing at conferences:  It's 
information you need to know when something *goes wrong*.  Performance 
management tools let you find out not only *what* is happening, but *why*. 
 

'Good Performance' is a qualitative term:
- It's relatative only to 'Bad Performance'
- It depends on properly functioning hardware and software
- It depends on sufficient raw resources
- It depends on the correct allocation of those resources

The CP INDICATE command really can't provide you with warm fuzzies on all 
those points.

Alan Altmark
z/VM Development
IBM Endicott

Reply via email to