On 24/08/2006 at 14:40 CET Bill Bitner <[EMAIL PROTECTED]> wrote:
> INDICATE can be misleading. The Toolkit and ESAMON reports should be
> valid. See http://www.vm.ibm.com/perf/tips/lparinfo.html for some
> background.
> Bill Bitner - VM Performance Evaluation - IBM Endicott - 607-429-3286
I accept that INDICATE can be misleading. However, to quote from the attached document,
|
which would lead me to believe that it bore some relation to the overall load on the processors being shared by the partition - and this seemed to be the case before be changed the processor assignment to the lesser partitions.
My concern is purely the hard cap that VM2 seems to have run into. The whole feel of the thing is that VM2 now has a cap where it did not before.
Since the changes a CP IND of 100% matching 64% of the 7 processors assigned indicates that VM2 had reached a hard limit - and there were signs of a master processor overload on VM2 (as you would expect). Looking at the system now when it is not busy I see the following :-
CP Ind on VM2 - 85% of 7 processors
Use of the real processors by VM2 (ESAMON) - 54.4%
If you extrapolate these 2 figures you get back to 100% = 64% of the real system which is what we saw yesterday. I cannot explain where this 64% figure comes from because the VM2 share is 52% so 64% is 123% of its share.
The practical effect of this is that when VM2 got heavily loaded on a full system before it would slow down but it would always 'steal' a little more from other partitions so that it did not hit a hard wall - which is what seems to be happening now. I just cannot see how taking virtual processors from ANOTHER partition could adversely affect this one.
