Subject: | |
From: | |
Reply To: | |
Date: | Tue, 18 Mar 1997 16:24:54 -0500 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
Chip,
System tuning is somewhat of a black art, but I have to agree with you
that the current parameters set on your 995-500 seem out of whack to
me too. Your current environment has all three queues embedded into
one another rather than overlapping one another or entirely separate from
one another. And with the queue quantum all set at the same low value, you
are not getting any performance benefit for the different types (i.e.
sessions or batch) of work that the computer is processing.
This setup could be a major cause of your loader problems and erratic
response times. With such low quantum values, your computer is probably
thrashing processes in and out of memory with very little time left to do
any useful work. Furthermore, I doubt that HP "recommended" your current
system settings since they represent everything you are NOT supposed to do
with system queue settings as I remember this subject being taught in their
System Manager's course.
My 959KS/300 system is tuned using the values below:
------QUANTUM-------
QUEUE BASE LIMIT MIN MAX ACTUAL BOOST TIMESLICE
----- ---- ----- --- --- ------ ----- ---------
CQ 152 202 1 200 8 DECAY 200
DQ 198 212 200 200 200 OSC 200
EQ 206 224 500 500 500 DECAY 200
The above setup favors my 425-450 on-line users for interactive sessions in
a MRP II environment. But it also allows for timely batch performance in
the "DQ" since I have overlapped it a little with the "CQ" and I also
oscillated this queue. My "EQ" is basically "no-man's land", but it too is
overlapped a little with the "DQ". I use it for any job that I really
don't care how long it takes to complete. My users enjoy sub-second
response time more than 95% of the time regardless of workload mix using
the above settings on my machine.
I use the exact same settings on my recovery provider's computer which is a
995-400 and I haven't experienced any erratic performance problems with it
during my live recovery tests. As always, your mileage may vary!!
Just my thoughts on the subject,
John Hornberger
Sr. Systems Programmer / HP Platform Manager
General Signal Services
[log in to unmask]
______________________________ Original Message Follows ______________________
Subject: Seeking opinions on TUNEing
Author: ChipDorm <[log in to unmask]> at Internet2
Date: 3/12/97 12:43 AM
I need some opinions. During my career I have done some
processor schedule tuning on single processor 3000s. The
current machine I am on has the tuning set this way:
------QUANTUM-------
QUEUE BASE LIMIT MIN MAX ACTUAL BOOST TIMESLICE
----- ---- ----- --- --- ------ ----- ---------
CQ 152 202 30 30 30 DECAY 200
DQ 170 202 30 30 30 DECAY 200
EQ 158 202 30 30 30 DECAY 200
HPCPUNAME = SERIES 995-500
I don't pretend to be any sort of expert here but this really
looks odd to me. I have no access to be able to alter the
tuning, and the people who are responsible for this machine
say that this is the setting that HP recommended. {8-0
The response from the 3000 is at times poor, having long loader
times ( > 30 secs) and even longer file access problems. There
are typically 650 user processes at any given time and most are
active.
Does this processor tuning look OK to y'all. If so why or why
not?
Oddly, GLANCE does not seem to show that there are any problems.
Is it possible that GLANCE has a problem in a multiple-processor
machine?
Chip "I-Hate-Internal-Politics" Dorman
|
|
|