HP3000-L Archives

February 1998, Week 3

HP3000-L@RAVEN.UTC.EDU

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
GeorgeStachnik <[log in to unmask]>
Reply To:
Date:
Tue, 17 Feb 1998 11:38:31 -0500
Content-Type:
text/plain
Parts/Attachments:
text/plain (190 lines)
One of the great things about system performance is that understanding
it really
boils down to a very small number of very basic principles.  The
principles don't
change - they hold on 4.0, 5.0 and (now) 6.0.  In fact, they hold on
UNIX, MPE
NT and every other OS I can think of.

'Way back in August of '96 there was a posting to this newsgroup from a
customer
who was upgrading from 4.0 to 5.0 and having problems similar to the
ones you
talked about in your posting.  I posted a response 'back then which
(thanks to
Deja News) I was able to find and repost now, with a few revisions.

The release numbers are different, but the principles are the same.  I
can't be
sure that they apply to your situation -- (don't know enough about it)
-- but
they're still worth repeating...

-geo

 ==============================================================================
=========


Subject:      Re: MPE/iX 4.0 to 5.0 or 5.5 [revised]
From:         George Stachnik <[log in to unmask]>
Date:         1996/08/28
Message-ID:   <[log in to unmask]>
Newsgroups:   comp.sys.hp.mpe

MPE/iX is a little different from other operating systems with respect
to performance.
With most OS's, whenever you add to the workload of a system,
performance get's a little
bit worse.  As you get close to the limit, each additional user begins
to "hurt" more
and more - but if you were to graph response time against the number of
users, you'd get
a curve that slopes up as the system gets busier, becoming steeper and
steeper as you
add more users.

By contrast, MPE/iX's performance curve tends to be relatively flat.
That is, whatever
response time you get with a small number of users will be pretty much
the same response
time you get with more users - at least up to a point.  The curve tends
to stay flat
until you get close to the limit.  Then it shoots up dramatically.

The charts below are crude, (you may have to adjust the font that your
newsreader uses to display them) but they make the point.  The
horizontal axis
represents the number of users logged on and working at any particular
time.
Logging more users onto your system moves you to the right.  The
vertical axis is
response time - as you move up, the response time gets worse (longer).
(Clearly,
these curves don't represent any particular system - they are just
generalizations).


                              |                                        |
                             +|                                       +|
                            ++|                                       +|
                          ++  |                                       +|
                        ++    |                                       +|
                    ++++      |                                      ++|
               +++++          |                                     ++ |
           ++++               |                                  ++++  |
     ++++++                   |                               +++++    |
+++++                         |      ++++++++++++++++++++++++++        |
------------------------------|     -----------------------------------|

Most Operating Systems                           MPE/iX


On most OS's, moving up the performance curve is like driving up a hill
that keeps
getting steeper until, at the limit, it's too steep to drive on.  On
MPE/iX, moving up
the performance curve is like driving on a flat road until you hit the
limit (the
dreaded "knee of the curve").  Then adding even one user can cause
performance to shoot
up like a skyrocket.  It is literally like running into a brick wall.

Now - what happens when you upgrade your operating system, (e.g. move
from 4.0 to 5.0,
or 5.0 to 6.0 or whatever)?   In most cases, the new O/S will be more
complex than
the old one because it has new features.  (You may not be using them,
but they're
in there nevertheless).  Of course there are exceptions - but if there's
a general
rule about operating systems, it's that new releases are more complex
and run
slower than old releases.  This is true of all OS's, not just MPE/iX.
(Remember
moving from Windows 3.1 to Win95?)

When you upgrade to a new, more complex release of MPE/iX it's like
shifting the
entire performance curve to the left by a notch or two.  What's that
going to do to
your performance?  The answer is (all together now) "It depends".

Take a second look at "other operating systems" graph above.  On other
operating
systems, (not on MPE/iX) updating to a more complex version of the O/S
is *always*
going to buoy you up a little bit.  The closer you are to the right hand
side of
the curve, the worse it's going to be, because that's where the curve is
the steepest.

Now look at the MPE/iX curve.  Suppose your system were not being pushed
to its
limits performance-wise.  That would put you on the flat part of the
curve.  Updating
to a new release might shift the curve to the left, but that would not
buoy you
up - and the whole update process would be a non-event.  Believe it or
not, that
is exactly what most MPE/ix customers report when they go to a new
release.  No
discernable change in performance.

On the other hand, if your system was being worked hard - i.e. if you
are already
close to the "knee" of the performance curve before you update to the
new version,
then upgrading is going send you shooting up very quickly.  In other
words, it
will have precisely the kind of impact that you describe in your posting
- longer
response times for online users, and longer run times for batch jobs.

Before you do the update, you might want to ask yourself these
questions:

*       Is performance ever erratic?  If you're near the "knee" of the
curve, then
        on busy days, performance will get much worse than it is on
so-so days.
*       Do you have access to performance measurement tools for the
3000?  These
        can be useful for figuring out where you are on the curve.
*       Do you have access to another system that you can use to test
your application
        on the new release?

One last point - everything that I've written in this message is a
generalization.  And
all generalizations are wrong (including this one).  :>  Anyway, good
luck with your update.



Alan Stowell wrote:
>
> Mark Gross  Says:
>
> I updated MPE/ix from 5.0 to 5.5 pp4 over the weekend. The process went
> fairly well and took about 6 hours. I was a bit displeased, however, to see
> my batch jobs taking up to 42% longer to complete, with the average being
> about 30%. YIKES! I expected to take a performance hit, but is this normal?
> We are running a 937LX with 128Meg of memory.
>
> Mark Gross
> Emerson Motion Control
> [log in to unmask]

--
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
-
George Stachnik                     VoiceMail (408) 447-4156
c/o Hewlett-Packard                 Fax:      (408) 447-4952
Installed Base Marketing Pgm's.     email     [log in to unmask]
19111 Pruneridge Ave | MS 44UL
Cupertino CA 95014

Watch the Hewlett-Packard Technology Closeup Interactive TV Broadcasts:
See http://www.hp.com/go/tapes for details on videotapes.

ATOM RSS1 RSS2