Re: WebSphere Performance Advisor

[Simon Coulter <shc@xxxxxxxxxxxxxxxxx>]

On 17/02/2009, at 3:14 AM, Joe Sam Shirah wrote:

>     Thanks to both you and Simon.  I was hoping that there was  
> something
> that would give the base WebSphere activity level or threads given  
> some
> load.  Then I could add an educated guess for the number of users.

See the Performance Capabilities Reference--Chapter 6 covers web  
server and WebSphere performance. Also mentions the Systems Workload  
Estimator but that's usually for sizing an appropriate system rather  
than tuning performance on a given system.


> I will look into the Web Performance Advisor.

Didn't know about that tool.

>     I am monitoring the threads entry that QPRFADJ makes.  As a  
> fallback, if
> necessary, I will use the value differential when we move WAS from  
> *BASE.
> While I believe Simon's thread/job explanation, I think this is  
> like some
> other areas of the AS/400 in that I'm not sure that's the entire  
> story.  For
> example, QPRFADJ currently has things at a little over 300 for *BASE
> activity level.  The application supports around 2000 users, and  
> WAS all by
> itself runs many threads.  I realize the significance of *concurrent*
> threads, but still...

It is not necessary to have an activity-level for each active thread-- 
only for those actually doing work. Inactive threads (i.e. those  
waiting for something to happen such as I/O) can be paged out.  
Examine the transitions Active->Wait, Wait->Ineligible, and  
especially Active->Ineligible.

Also, is all your work occurring in *BASE? If so that's not a good  
idea. Better to separate workloads by performance profile (e.g.,  
batch, interactive, web, communications, etc.). *BASE should be left  
for the subsystem monitor jobs, some IBM jobs, plus as a pool of  
available storage for the Performance Adjuster to move storage into  
and out of.

If you've already moved workload out of *BASE then that will help  
explain the activity-level of 300. Note that the system is very good  
at managing main storage, as long as you have sufficient to avoid  
thrashing, so it's certainly possible that 2000 users could be  
supported by 300 activity-levels but I suspect you have work  
occurring in other pools.

Regards,
Simon Coulter.
--------------------------------------------------------------------
   FlyByNight Software         OS/400, i5/OS Technical Specialists

   http://www.flybynight.com.au/
   Phone: +61 2 6657 8251   Mobile: +61 0411 091 400        /"\
   Fax:   +61 2 6657 8251                                   \ /
                                                             X
                 ASCII Ribbon campaign against HTML E-Mail  / \
--------------------------------------------------------------------