RE: Performance of batch vs other batch and interactive -- MIDRANGE-L

Thanks for the performance 101 information.
I have changed the subsystem to it's own pool and only allocated a little
memory to it.
This seems to have fixed the initial problem of them taking up all the
system resources.

Now I will go back and start "fixing" the rest of this system setup. It's
sounds like from what you say our system is setup all wrong.

I prefer to make only a change or two and "let it ride" for a bit to see the
affect it has, as opposed to making a lot of changes and then encountering a
problem and not knowing which change caused the problem.

Thanks for all the info, it is greatly appreciated

John



-----Original Message-----
From: midrange-l-bounces@xxxxxxxxxxxx
[mailto:midrange-l-bounces@xxxxxxxxxxxx] On Behalf Of Clare Holtham
Sent: Friday, June 04, 2004 2:53 AM
To: Midrange Systems Technical Discussion
Subject: Re: Performance of batch vs other batch and interactive

Yes! 
The first principle of performance tuning is to separate the workloads.
Batch in its own pool (not in *Base), and interactive in its own pool. QCTL
controlling. If you have lots of ODBC etc you can move that out of *Base
also. Lots of important things happen in *Base anyway (file opens, subsystem
monitors, client access, etc etc). Check that you have 'Expert Cache'
enabled (WRKSYSSTS in Expert mode, then F11 and you should see *Calc not
*Fixed against the pools.) If not, change it to *Calc. Over time this will
monitor the workload in the pool and optimize it. If you have a mix of batch
and interactive or anything else, it cannot do this!
Having said this, I think you need to look at how you have these jobs set up
anyway. You have given them quite a short timeslice - maybe a longer one
would work better.
And IBM still ships the QINTER class with a timeslice of 2000 (2 secs) which
is way too long, you should change this to around 200 or 300 (you can
calculate what it should be, but this is a guesstimate) which will give you
better throughput in interactive, which will help generally too.
And if you have these 10 batch jobs in their own memory pool, you can always
'constrain' them be reducing the memory. (Don't forget to turn off the
automatic tuning though.)
I also like the suggestion of having a way to make them wait when you need
to!
Hope this helps,

cheers,

Clare

Clare Holtham
Director, Small Blue Ltd - Archiving for BPCS
Web: www.smallblue.co.uk  
IBM Certified AS/400 Systems Professional 
E-Mail: Clare.Holtham@xxxxxxxxxxxxxx
Mobile: +44 (0)7960 665958
  ----- Original Message ----- 
  From: Richard Allen 
  To: 'Midrange Systems Technical Discussion' 
  Sent: Thursday, June 03, 2004 8:46 PM
  Subject: RE: Performance of batch vs other batch and interactive


  Here is some additional information, 
  We have a model 720  with two processors 
  8g memory
  400g+ DASD (Running at about 70% capacity)

  Each of the 10 jobs is doing it's own tasks and barely keeps up with the
  Data Queue feeding it, so I don't think we can combine them into less
jobs.

  The 10 batch jobs are running in Subsystem pool 2 (same as the interactive
  and other batch jobs)

  Is this the problem? Should it be running in it's own pool with a smaller
  amount of memory?



  -----Original Message-----
  From: midrange-l-bounces@xxxxxxxxxxxx
  [mailto:midrange-l-bounces@xxxxxxxxxxxx] On Behalf Of darren@xxxxxxxxx
  Sent: Thursday, June 03, 2004 9:40 AM
  To: midrange-l@xxxxxxxxxxxx
  Subject: Re: Performance of batch vs other batch and interactive





  I'll preface my advice first, with a disclaimer.  I'm a pure programmer
and
  not a system administrator.  That being said, I would try the following
  things.  Also, you're asking for some pretty specific information without
  specific information about your system.  Your fellow forum members might
  have more ideas if you describe you setup a little more (# of processors,
  subsystem setup, archive method, etc.)

  1. If all 10 jobs are doing the same thing, then reduce the number of
  running jobs.  This would probably be the most significant thing to do.
  This would be especially true on a multiple processor machine, if this
  applies.  10 jobs could occupy a significant portion of a 12-way
processor,
  while one job might only use 1 processor max.
  2. I think I'd increase the jobs run time from 1000 to at least 2000 ms.
  I'm guessing that the job can get more meaningful stuff done with the
  greater time, and the shorter times might tend to increase the overall
  overhead as the thread starts and stops.
  3. Perhaps adjust the subsystems memory usage.  I'm not sure if more or
  less would give you better results.
  4. Can the actual process performance be improved?  If you're using
  Infoprint Server to archive, make sure you aren't embedding fonts and such
  in the PDF file.

  > What I want is for the 10 batch jobs running in their own subsystem to
  take
  > the lowest priority when there is high CPU utilization caused from users
  > running interactive jobs and submitting their own jobs to batch.
  >
  > Any suggestions?
  > Thanks
  > John

  --
  This is the Midrange Systems Technical Discussion (MIDRANGE-L) mailing
list
  To post a message email: MIDRANGE-L@xxxxxxxxxxxx
  To subscribe, unsubscribe, or change list options,
  visit: http://lists.midrange.com/mailman/listinfo/midrange-l
  or email: MIDRANGE-L-request@xxxxxxxxxxxx
  Before posting, please take a moment to review the archives
  at http://archive.midrange.com/midrange-l.



  --
  This is the Midrange Systems Technical Discussion (MIDRANGE-L) mailing
list
  To post a message email: MIDRANGE-L@xxxxxxxxxxxx
  To subscribe, unsubscribe, or change list options,
  visit: http://lists.midrange.com/mailman/listinfo/midrange-l
  or email: MIDRANGE-L-request@xxxxxxxxxxxx
  Before posting, please take a moment to review the archives
  at http://archive.midrange.com/midrange-l.
--
This is the Midrange Systems Technical Discussion (MIDRANGE-L) mailing list
To post a message email: MIDRANGE-L@xxxxxxxxxxxx
To subscribe, unsubscribe, or change list options,
visit: http://lists.midrange.com/mailman/listinfo/midrange-l
or email: MIDRANGE-L-request@xxxxxxxxxxxx
Before posting, please take a moment to review the archives
at http://archive.midrange.com/midrange-l.