"Is there another operating system that can operate anywhere near 100%
busy?"
My counter question would be "Is there another one that bills itself as
able to, and charges for hardware and software accordingly?"

The less wonderful to which I refer is most noticeable when we are at
100%. And we are. Regularly.

Here's a more obvious example:

Amongst other things, we run a bunch of APPC jobs. They are for offline
reporting - one of those 1..n at a time deals depending on how much work
we have waiting. They all run at a level right after the TP systems, so
they do get some CP. They all also run at the same level, so they should
in theory all get the same CP.

A controller to these things acts like a traffic cop - it directs work to
a free thread and starts more (up to a limit) if there's a queue. After a
few minutes of inactivity it closes them down gracefully. This isn't new
code - we've been doing this for >15 years and it works just fine.

Over the years, with changing workloads we've put more and more work
through this application. We've noticed that when the system is busy, WLM
is loathe to actually switch in and run a new ASCH. We get as far as
having it init, but then it just sits there doing nothing.

The more we have - and it all goes pear shaped around the 230 concurrent
mark in this LPAR in our installation - the worse the problem is. We've
had these things sitting there for 10 minutes doing nothing, while their
previously started peers churn away quite merrily. We've learned to live
with it (or more accurately our clients have) since we hard cap at 200 or
so now. In a completely fair system I'd expect to have all of them each
getting a little CP, not the few we already have running and the rest
doing nothing.

Now if there's a fix we missed, I'm all ears.
This is, however, a representative example of why we're nervous of address
space proliferation...






From:   Peter Relson <[email protected]>
To:     [email protected]
Date:   04/12/2011 07:51 AM
Subject:        Re: Address space proliferation
Sent by:        IBM Mainframe Assembler List
<[email protected]>



>It's this quote that's driving this problem
>"z/OS handles many address spaces very well."
>
>It's far less wonderful at handling them (at least in our experience)
when
>the machine is busy.

On the contrary, I assert that it handles them extraordinarily well. Is
there another operating system that can operate anywhere near 100% busy?

If your reason for thinking "less wonderful" is because you are not
getting enough CPU time when the system is busy, then consider blaming the
WLM settings that your site set which have deemed your application not
worthy of a larger share.

>I agree that limiting the number of dispatchable units of work (TCB's) to
>the number of CPs is not particularly useful.

You may agree, but I believe that you are mostly incorrect.  Once it is
asserted that the tasks do not wait (which I believe was asserted), you
cannot get much more work out of 11 TCBs than 10 TCBs if there are only 10
CPUs.
You certainly cannot have them all run at the same time.  Yes, if one task
is waiting for a page fault to be resolved, you could have another one do
some work. But that all presumes that you are the only user of the LPAR,
otherwise other work will be competing with you to run.

Peter Relson
z/OS Core Technology Design

Reply via email to