Hi all
This is a snapshot of a paper I wrote a couple of years ago.
May be it can help ...
Unfortunately I could not attach the graph because of List rules but results
using SMF 100 are just a little bit lower than using SMF 30 OR 72.
Best regards
Fabio
DB2 overhead for MSTR, DBM1 and IRLM can be easily evaluated using SMF 30, 100
or 72
records.
a) Using SMF 30 interval records (subtype 2 and 3); you have to select the
records belonging
to MSTR, DBM1 and IRLM address spaces and sum the CPU time provided in the
SMF30ICS, SMF30ICU, SMF30HPT, SMF30IIP, SMF30RCT, SMF30CPS, SMF30CPT
fields;
b) Using SMF 100; you have to use the QWSAEJST and QWSASRBT fields. A section
for
each DB2 system address space is provided, so to get MSTR, DBM1 and IRLM
overhead
you have to sum the values corresponding to all of them2. Remember that these
counters
have been accumulated since DB2 was last started. So a de-accumulation step is
required to
get the numbers relative to the analyzed period of time.
c) Using SMF 72 records requires a preliminary assignment of DB2 system address
spaces to
specific WLM service or report class. Then you have to:
1. select the records belonging to these classes;
2. normalize zAAP and zIIP service units, provided in the R723CIFA and R723CSUP
fields, to standard CPUs speed multiplying respectively by the R723NFFI and
R723NFFS coefficient (they normally have the same value) and dividing by 256;
3. subtract normalized zAAP and zIIP service units from the service units
provided in
the R723CCPU field;
4. convert TCB and SRB service units values in R723CCPU and R723CSRB to CPU
seconds multiplying by the system service units per second value and dividing
by the
TCB and SRB coefficients (provided in the R723MCPU and R723MSRB fields);
5. sum the values obtained in the previous step to the values provided in the
R723CIIT,
R723CHST and R723CRCT fields.
We applied all the methods described above and we calculated the DB2 System
Address Spaces
overhead using the following variables:
• AS100OVH, based on SMF 100 records;
• AS072OVH, based on SMF 72 records;
• AS030OVH, based on SMF 30 interval records.
The graph in Figure 1 allows us to compare the results we obtained. Only the
“prime shift” hours,
from 8am to 5pm, are presented.
+
+ Fabio Massimo Ottaviani
+ EPV Technologies Technical Director
+ Skype: fabio.massimo.ottaviani
+ Mobile: +393406168088
+
+ IT Cost under Control
+ www.epvtech.com
+
Please consider the environment - do you really need to print this email?
-Messaggio originale-
Da: IBM Mainframe Discussion List [mailto:IBM-MAIN@LISTSERV.UA.EDU] Per conto
di Barry Merrill
Inviato: giovedì 8 agosto 2013 19:27
A: IBM-MAIN@LISTSERV.UA.EDU
Oggetto: Re: Differences between CPU values from SMF30 and DB2 stats
In general, there will be very little CPU time recorded in the DB2 Address
Spaces, except for the DDR address space, because CPU time consumed in DB2 is
recorded in the address space OF THE CALLER, i.e., the Batch Job or the CICS
region that called DB2, so the SMF 101 (DB2ACCT) data is the only safe source
of who is consuming DB2 CPU time.
Barry
Herbert W. Barry Merrill, PhD
President-Programmer
MXG Software
Merrill Consultants
10717 Cromwell Drive
Dallas, TX 75229
mailto:ba...@mxg.com ba...@mxg.com
http://www.mxg.com http://www.mxg.com - FAQ has Most Answers
mailto:ad...@mxg.com ad...@mxg.com - invoices/PO/Payment
mailto:supp...@mxg.com supp...@mxg.com- technical
tel: 214 351 1966 - expect slow reply, use email
fax: 214 350 3694 - prefer email, still works
-Original Message-
From: IBM Mainframe Discussion List [ mailto:IBM-MAIN@LISTSERV.UA.EDU
mailto:IBM-MAIN@LISTSERV.UA.EDU] On Behalf Of Mike Bell
Sent: Thursday, August 08, 2013 12:21 PM
To: mailto:IBM-MAIN@LISTSERV.UA.EDU IBM-MAIN@LISTSERV.UA.EDU
Subject: Re: Differences between CPU values from SMF30 and DB2 stats
There is a basic issue with DB2 statistics for CPU - they start with the first
SQL statement. Everything that happens before that is not recorded in the DB2
cpu times. This is WAD. There used to be some presentations on what was
included in DB2 cpu and why. I haven't looked for them in a long time. Maybe
someone on DB2 listserv has a better memory or keeps better notes.
Mike
On Thu, Aug 8, 2013 at 8:39 AM, Charles Mills mailto:charl...@mcn.org
charl...@mcn.org wrote:
Do you know that there is a DB2-specific mailing list where lots of
serious DB2 performance heavyweights hang out?
I believe the enrollment process is to go to http://www.idug.org
www.idug.org and join IDUG.
Charles
-Original Message-
From: IBM Mainframe Discussion List