Hi Barbara Jim, and others, thanks for your comments.
Just how did you arrive at the conclusion that elongated I/O times are the
source of your problem? How did you measure them?
This is the most frustrating thing, there is no measure that point to it, it's
all inference.
The RMF I/O response time measures roughly the same on each LPAR, as measured
by RMF.
I/O bound jobs take longer to run on the smaller LPAR when capped 2 -3 times
longer
This effect is seen even if the job is placed in a very high service class.
The overall I/O rate on the small LPAR reduces significantly when capped, ie
4000->2000 I/Os per sec.
I've read the Kathy Walsh presentation on short CPs, and it matches. The
mvsbusy/lparbusy increases to 4-5 on bad intervals, tying with the IO rate
drop. The issue appears to be the weights have not been changed as work has
migrated onto the smaller LPAR, and there are to many LCPs online. Both LPARs
have 11 LCPs online, with 13 physicals.
Our plan is to make manual adjustments to online CPs and weights for our next
monthly peak, and then implement IRD weight and CPU management after this. The
profile changes significantly overnight and on the first working day of the
month, so automatic management would be good. Does anyone have any experiences,
good or bad of IRD implementation?
With regard to CPENABLE, I still think this may be coming into play, but only
as a side effect. The top 4 LCPs handle 96% of the I/O interrupts due to
CPENABLE But if I add the LPAR busy times of these 4 CPs, the max online ime
that any one of these CP for this LPAR is 73%. It will be less, because their
busy times will overlap to some extent. (oring probabilities is harder than
anding them!).
I've pasted below an example RMF CPU interval, any further thoughts are most
welcome.
Joe
1 C P U A C T I V I T Y
PAGE
z/OS V1R11 SYSTEM ID SYSB START
02/01/2012-09.00.00 INTERVAL 000.30.01
RPT VERSION V1R11 RMF END
02/01/2012-09.30.02 CYCLE 1.000 SECONDS
-CPU 2094 CPC CAPACITY N/A SEQUENCE CODE 00000000000AEDEA
MODEL 713 CHANGE REASON=N/A HIPERDISPATCH=N/A
H/W MODEL S18
0---CPU--- ---------------- TIME % ---------------- LOG PROC --I/O
INTERRUPTS--
NUM TYPE ONLINE LPAR BUSY MVS BUSY PARKED SHARE % RATE
% VIA TPI
0 CP 100.00 18.42 89.45 ------ 17.0 37.68
75.33
1 CP 100.00 18.44 89.14 ------ 17.0 37.29
76.70
2 CP 100.00 18.43 88.79 ------ 17.0 37.51
74.81
3 CP 100.00 18.47 89.09 ------ 17.0 37.11
74.64
4 CP 100.00 18.40 88.82 ------ 17.0 36.42
74.49
5 CP 100.00 18.37 88.14 ------ 17.0 35.60
74.83
6 CP 100.00 18.40 88.61 ------ 17.0 34.94
75.27
7 CP 100.00 18.36 88.64 ------ 17.0 737.0
13.49
8 CP 100.00 18.35 88.28 ------ 17.0 834.1
14.99
9 CP 100.00 18.30 86.94 ------ 17.0 823.6
14.64
A CP 100.00 18.32 87.53 ------ 17.0 829.2
14.99
TOTAL/AVERAGE 18.39 88.49 187.0 3480
19.03
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
