Re: How much memory is currently paged?

[Jim Oberholtzer <midrangel@xxxxxxxxxxxxxxxxx>]

Evan,

You have just touched on the black art portion of tuning.  The real 
answer is it depends, on how the application is behaving, what else is 
running on the system, etc.  As a general rule I like to see faults <20% 
of the page rates if it is the database side.  By definition data is 
transient and there will be some faults during I/O operations.  If the 
application is almost write only then I would expect that ratio to be 
quite a bit better maybe <5%.  Paging in the Database columns is real 
work being done in DB/2 so a high page rate there is indicative of quite 
a bit of I/O.  Usually a good thing.

On the program side (Non-DB) keep faults/pages in the machine pool and 
*BASE as close to zero as possible.  .1 <> .8 is OK, but over that your 
starting to thrash IBM i and the licensed program products.  On the P7 
boxes the numbers will be a bit higher but the goal is to minimize 
them.  For Domino &WAS and any other JAVA based applications they are 
going to page/fault like mad so try to keep them at <20% of the total.  
Remember that garbage collection in JAVA is going to clobber a bunch of 
paged out memory so you'll see activity just for that.

The best way to see what your applications are doing  is to use IBMs 
Performance Tools (all the parts which include iDoctor and Job Watcher 
now) and MPGs Performance Navigator.  Those are the two sets of tooling 
I use to diagnose application performance problems and suggest 
corrections to the code.   Look for job wait states.  A wait state will 
cause paging / faulting since a waiting program will get flushed from 
main store quickly, then once the wait is satisfied, the system has to 
bring it all back in again, more paging/faulting.....

Also keep your DB/2 indexes up to date and you'll help the system out a 
bit as well.

Jim Oberholtzer
Chief Technical Architect
Agile Technology Architects


On 8/14/2012 6:08 PM, Evan Harris wrote:
> So what kind of ratio do you look for to between paging and faults -
> do you have a rule of thumb as to what % of faults is acceptable ?
>
> On Wed, Aug 15, 2012 at 3:29 AM, DrFranken<midrange@xxxxxxxxxxxx>  wrote:
> > >  Depends on the system (CPUs, total memory) the faulting numbers here
> > >  could be acceptable but they are significant. Pool 2 (*BASE) is showing
> > >  more than one in four pages as a fault that's not a great ratio. Pool 5
> > >  is very poor with about 80% of pages faulting. Need more meory there
> > >  almost certainly. Pool 4 is very good with 0.03% of pages faulting and
> > >  that's with a LOT of paging!  The machine pool ration is poor also with
> > >  almost seven of eight pages faulting but the numbers are pretty low
> > >  overall there so not a huge problem.
> > >
> > >  If it was me, I would be investigating more memory for this system or
> > >  potentially fewer jobs running in pool 2 and 5.
> > >
> > >  Of course additional performance work with the iDoctor tools or MPG's
> > >  tools will help get more refined answers.
> > >
> > >  I don't believe QTEMP is in there, that space is tracked differently as
> > >  I recall.
> > >
> > >        - Larry "DrFranken" Bolhuis
> > >
>
> -- Regards Evan Harris http://www.auctionitis.co.nz
> --