SMF42 subtype 6 interpretation

[Ducky Duc]

Hi all ,

As a DB2 specialist , i have reports showing a wait for io time in some programs. I did a SMF42 subtype 6 reports and , effectively, my datasets are in the top of the list , cut and pasted below from an Excel sheet.
However , don't know if my calculation of the average response time of 9 ; which corresponds to 9 X 128ms = 1152ms , from the manual is correct. And sorry for the dummy question , what is a "correct" average response time , a few ms i suppose ...

Thank you
Duc

DATASET VOLSER RESPTIME IOCOUNT AVGCONCT AVGPEND AVGDISC
DB2XXF0.DSNDBD.F4DB002.F4SS102.I0001.A001 XDBDO0 22.875 5379 5.187 0.125 17.250
DB2XXF0.DSNDBD.F3DB001.F3SS039.I0001.A001 XDBDM5 21.625 1141 16.750 0.125 4.437
DB2XXF0.DSNDBD.F3DB002.F3SS113.I0001.A001 XDBD8A 14.062 7489 1.875 0.125 11.625
DB2XXF0.DSNDBD.F3DB002.IF3TADR1.I0001.A001 XDBI16 12.875 2057 1.750 0.125 10.750
DB2XXF0.DSNDBD.F3DB001.F3SS037.I0001.A001 XDBDH3 9.562 3946 6.125 0.000 3.187
DB2XXF0.DSNDBD.F3DB001.IF3T1BGJ.I0001.A001 XDBI08 9.562 2038 5.625 0.000 3.562
DB2XXF0.DSNDBD.F3DB006.F3PS006.I0001.A003 XDBD53 9.187 1244 5.875 0.125 2.937
DB2XXF0.DSNDBD.F3DB006.F3PS006.I0001.A007 XDBDC0 9.062 1203 5.750 0.125 2.937
DB2XXF0.DSNDBD.F3DB006.F3PS006.I0001.A008 XDBDF6 9.062 1321 6.000 0.125 2.687
DB2XXF0.DSNDBD.F3DB006.F3PS006.I0001.A009 XDBDKD 9.062 1321 5.750 0.125 3.062
DB2XXF0.DSNDBD.F3DB006.F3PS006.I0001.A010 XDBDI2 9.062 1328 6.250 0.125 2.375
DB2XXF0.DSNDBD.F3DB001.IF3TCCTD.I0001.A001 XDBI4C 9.062 1057 5.625 0.000 3.187
DB2XXF0.DSNDBD.F5DB001.F5SS012.I0001.A001 XDBDG0 9.062 1926 7.000 0.125 1.625

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to list...@listserv.ua.edu with the message: INFO IBM-MAIN

[Elardus Engelbrecht]

Ducky Duc wrote:

>However , don't know if my calculation of the average response time of 9 ; which corresponds to 9 X 128ms = 1152ms , from the manual is correct. And sorry for the dummy question , what is a "correct" average response time , a few ms i suppose ...

Could you be kind to post the SMF data description here? For example in the SMF book S42DSIOR is 'Average response time' and S42DSIOP is 'Average I/O pending time', etc.

This is to help others to properly answer your question.

For my side, I would also look beside the IOCOUNT (what is it named in SMF book?), also to number of blocks and any delays details. I'm more concerned about eventual application response time, but if that is suffering from those datasets, yes, I would drill down there on those devices too.

About average response time, what devices type do you report on?

What is your RMF saying about those device delays if any?

Groete / Greetings
Elardus Engelbrecht

[DASDBILL2]

Has anyone complained yet about his response time's being too long?


Bill Fairchild
Franklin, TN

“Political language is designed to make lies sound truthful and murder acceptable, and to give the appearance of solidity to pure wind.” [George Orwell]

[Ducky Duc]

I guess that the value of 8 for S42DSIOR (Average Response Time) , i.e 128ms X 8 = > 1 second , is not good at all ....

[DASDBILL2]

I strongly suspect, without reading the doc, that "ms" should be microseconds instead of milliseconds.  I suspect this because of the multiplier 128.  I advise reading the SMF doc very closely.  If it says the value is a number of "timer units", then you have to find out what a timer unit means.  This is defined in the Principles of Operations as 128 microseconds.

Bill Fairchild
Franklin, TN

“Political language is designed to make lies sound truthful and murder acceptable, and to give the appearance of solidity to pure wind.” [George Orwell]

----- Original Message -----
From: "Ducky Duc" <ndt...@GMAIL.COM>
To: IBM-...@LISTSERV.UA.EDU

[Mackenzie, Bruce]

In the SMF manual above the Data Set I/O Statistics Section for the Subtype 6, it does say

"Note: I/O response and service time components are recorded in multiples of 128 micro-seconds."

So XDBDO0 has an average response time of 22.875ms of which 5ms is connect time and 17ms disconnect time.

The high connect time would indicate the reading (or possibly the writing) of multiple pages at a time (index or table space scan). The disconnect time of 17ms could also be considered high. Disconnect time is normally from random reads or if you are synchronously mirroring these devices via PPRC it can be from writes. So there is not a "correct" average response time, as it depends on the type of I/O being executed and your environment.

Typically the connect time for a single 4KB data page either read or write should be under 1ms for a modern control unit with a 2Gb FICON channel. A non-mirrored write will generally be a cache hit and therefor will not have any disconnect time. A mirrored write will have some disconnect time based on the size of the block, the speed of the link and the distance between the primary and secondary. A random read will have some disconnect time as the control unit pulls the page from the disk, and that time would depend on how busy the control unit / raid-rank was at the time.

Thanks

-----Original Message-----
From: IBM Mainframe Discussion List [mailto:IBM-...@LISTSERV.UA.EDU] On Behalf Of Ducky Duc
Sent: Thursday, April 18, 2013 9:54 AM
To: IBM-...@LISTSERV.UA.EDU
Subject: SMF42 subtype 6 interpretation

NOTICE: Confidential message which may be privileged. Unauthorized use/disclosure prohibited. If received in error, please go to www.td.com/legal for instructions.
AVIS : Message confidentiel dont le contenu peut être privilégié. Utilisation/divulgation interdites sans permission. Si reçu par erreur, prière d'aller au www.td.com/francais/avis_juridique pour des instructions.

[Martin Packer]

There IS some information on reads versus write in 42-6. I'm hazy on the
details but it could be useful with the "what is Disc?" question.

Cheers, Martin

Martin Packer,
zChampion, Principal Systems Investigator,
Worldwide Banking Center of Excellence, IBM

+44-7802-245-584

email: martin...@uk.ibm.com

Twitter / Facebook IDs: MartinPacker
Blog:
https://www.ibm.com/developerworks/mydeveloperworks/blogs/MartinPacker

From: "Mackenzie, Bruce" <Bruce.M...@TD.COM>
To: IBM-...@listserv.ua.edu,
Date: 04/19/2013 01:27 PM
Subject: Re: SMF42 subtype 6 interpretation

AVIS : Message confidentiel dont le contenu peut �tre privil�gi�.
Utilisation/divulgation interdites sans permission. Si re�u par erreur,
pri�re d'aller au www.td.com/francais/avis_juridique pour des

instructions.

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to list...@listserv.ua.edu with the message: INFO IBM-MAIN

Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number
741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU

[Shmuel Metz , Seymour J.]

In
<461643955.1081520.1366...@sz0127a.westchester.pa.mail.comcast.net>,
on 04/18/2013

at 06:03 PM, DASDBILL2 <dasd...@COMCAST.NET> said:

>I strongly suspect, without reading the doc, that "ms" should be

>microseconds instead of milliseconds.ￜ I suspect this because of
>the multiplier 128.ￜ I advise reading the SMF doc very closely.ￜ

>If it says the value is a number of "timer units", then you have

>to find out what a timer unit means.ￜ This is defined in the

>Principles of Operations as 128 microseconds.

No; PoOps does not define timer units; 128 microseconds is the
granularity of the channel measurements. The term timer unit is
defined as 26.04166 microseconds in, e.g., z/OS MVS Programming:
Assembler Services Reference, Volume 2 (IAR-XCT), SA22-7607-15. That
value is derived from the updating of bit 30 in the old S/360 Interval
Timer.

--
Shmuel (Seymour J.) Metz, SysProg and JOAT
Atid/2 <http://patriot.net/~shmuel>
We don't care. We don't have to care, we're Congress.
(S877: The Shut up and Eat Your spam act of 2003)

[Al Sherkow]

Yes the times are in uses of 128 microseconds.

If you are a member of CMG or are willing to register on their website you can find my 1998 paper "DOES YOUR INSTALLATION HAVE HALF-SECOND I/O RESPONSE TIME? (A Closer Examination of Disconnect Time)". I believe that paper first explored how to use the SMF 42-6 data to better understand response time for cache hits and cache misses.

Hard to believe that was 15 years ago that I did that work.


Al Sherkow

[Martin Packer]

Hi Al! Did it take into account Sync Rem Copy (PPRC and the like)? I don't
even remember if PPRC had been released back then.

Cheers, Martin

Martin Packer,
zChampion, Principal Systems Investigator,
Worldwide Banking Center of Excellence, IBM

+44-7802-245-584

email: martin...@uk.ibm.com

Twitter / Facebook IDs: MartinPacker
Blog:
https://www.ibm.com/developerworks/mydeveloperworks/blogs/MartinPacker

From: Al Sherkow <a...@SHERKOW.COM>
To: IBM-...@listserv.ua.edu,
Date: 04/21/2013 07:50 PM
Subject: Re: SMF42 subtype 6 interpretation
Sent by: IBM Mainframe Discussion List <IBM-...@listserv.ua.edu>

Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number
741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU

[Al Sherkow]

The examples did not use anything exotic like that. This was "direct" attached DASD and the write misses were occasionally half a second and that showed up in the SMF42-6 data as that records average and maximum service times.