Ted,
as Thomas Kern wrote, the graphics work just as intended: I liked (and still
prefer) the idea of automatically adapted scales when using the tool for
performance analysis ..
But I agree that for capacity planning purposes or general statistics a
fixed scale is often better. The clean solution would obviously be to allow
telling PerfKit about the preferred scale, but there's no way to do so now,
and you'll have to convince Endicott to implement it. In the absence of that
possibility you can try different ways to achieve a similar result:
A) My approach would have been to just plot the WHOLE period you're
interested in. That way you'll always have a good graph for the day,
and have a good chance of seeing the same scale also on other days.
And if it's for statistical purposes the whole day is probably better
anyway.
If you really need graphics for shorter periods the following may help.
B) Plot both the variables CPU and %WT, and specify 'cumulative' display:
Since the total of the two values should be a constant you should end
up
with a fixed scale.
The drawback is that you cannot include in the same output other
variables
whose value should not be cumulated, too (such as IO/S).
C) Plot CPU, IO/S and %WT, non-cumulative.
Since PerfKit will try to use the same scale for similar variables
(with
the same unit type, in this case '%CPU') the CPU and %WT values
should be plotted in the same scale. Will probably achieve the fixed
scale you want, and you can still include I/O as well.
The drawback is that you end up plotting the %WT variable that you
don't
really want.
D) You can also try to
- define a fixed 'user variable' with the max. scale to be used, and a
unit
type of '%CPU'
- plot this user variable together with 'CPU'.
Since PerfKit will try to use the same scale for variables of equal
units
the large (user) variable should determine the scale.
However, I'm not really sure whether this will work: it may depend on
the internal representation of the values (e.g. number of decimals),
and
then the user variable could end up being not quite the same. I just
don't
remember (and I doubt anybody ever tried it) ..
BTW: Do you really use the primitive 3270-mode PLOT screen FCX105?
I had assumed that by now everybody would be using the Web
interface
to create real graphics.
Eginhard Jaeger
,----- Original Message -----
From: "Ted Lewis" <[EMAIL PROTECTED]>
To: <[email protected]>
Sent: Friday, August 11, 2006 10:25 PM
Subject: IBM z/VM 5.2 Performance Toolkit
In working with IBM Performance Toolkit (z/VM5.2).
Screen: FCX105 (graphing a HISTLOG file).
I am plotting CPU (% total CPU) and I/O (SSCH/RSCH rate).
Does anyone know of a way to maintain a "fixed" scale value for the "Y"
axis in this report (FCX105). It currently changes based on the maximum
usage for the reporting period being viewed.
That is (we are trying to plot usage on a 24 hour cycle):
From 00:00-04:00 "Y" scale = 60 (lower usage)
From 04:00-08:00 "Y" scale = 200 (higher usage)
From 08:00-12:00 "Y" scale = 100, and so forth.
This is confusing - I prefer to have a fixed "Y" axis value based on my
input or a standard (ie. always show a scale based on CPU 200 and I/O
200). Its difficult to compare values when the scale value changes. I
questioned IBM and they do not know of a way (or do not think it
worthwhile
to investigate and develop a way). There is no mention of how to
set/change the "Y" scale in IBM's Performance Toolkit manual - it actually
states that the "Y" scale value changes based on maximum usage reported
for
the time period being reviewed.
Any suggestions/help would be appreciated. Thanks, Ted
