On Jan 24, 2008, at 10:08 AM, Ron Hawkins wrote:
ED,
I've have never observed SRB time to vary so much across MVS
versions and
releases, but I started with MVS/XA, Things may have been different in
earlier versions. However, if the address space is burning the SRB
CPU time
then the user must still be responsible for the cost. If you ignore
SRB time
all you will do is adjust the charge for TCB time to account for a
larger
uncaptured, and some other application will be funding the SRB time
of the
IO intensive applications.
The same variability can happen in TCB time, but you don't suggest
ignoring
this. In fact I've observed this in the G4 when IBM put the
compression
instructions into macrocode and the CPU time for jobs using SMS
compression
went through the roof. It did not come down again until the G6.
This was a
variation in CPU time of 100s of percent. Your example for SRB time
suggests
that this is a basis for ignoring TCB time.
Ron
Ron,
Interesting. My numbers were off the top of my head as I said that I
did not have a report in front of me. My memory (subject to dropped
bits) indicated something close to what I had written. Of course when
I looked back at the report and doing research after a minor
complaint by a user was that there wasn't any CPU change or anything
else that stood out in the change log (or my memory) of anything else
changing during the time in question. We had a pretty stable
environment (maybe to stable) except for MVS releases. I had marching
orders to get us up to pretty recent levels as we had plans to change
our disks from 3380's to 3390's and we were *NOT* fcs of the 3390's
but were after the the first day . The change we had to go through to
leap through different releases kind of quickly and it took a toll on
me with SMPe and all the maintenance that needed to be put on to get
to the point where it was doable.
I do remember that I had tried to push management to use the QCM
measurements as they were actually repeatable. I was getting a
little peeved that the customer was complaining about such small
differences (if they had been larger I would have agreed with him)
but this was semi not for profit company. This really meant we had
to return anything to the owners (option traders) any profit over
some percent unless we had asked to retain the money's for some up
coming extraordinary expense. In any case management refused to let
me QCM numbers (never understood why) and I could not figure out why
there was the variance. CPU time was a little variable depending on
trade volume. But seemed to be consistent and the users were OK but
the SRB time was less than consistent. Mind you it was never *HIGH*
just varied. by say 10 percent.
After we got up to speed on the software then things started to
change CPU wise and thats when the CPU charge back got interesting. I
got out of the responsibility as they hired a professional
performance tuner and he took care of the stuff. I fielded questions
from him like what release was on this or that day or other type of
questions but I lost track of the issue in any case.
Ed
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
