I know a little something about ABO (and a little less than that about 
COBOL V5 Optimisation) and I agree with your comments. They make perfect 
sense (and no I've not been involved in this situation).

I recently cautioned an account team to guide their customer thus. I think 
it's fine to share with y'all...



Take a CICS region that talks to DB2 (and we can detect this connection 
from SMF 30). Suppose it's 1 engine's worth (and no this is not a 
discussion about QR TCB limitations). We don't know from SMF 30 how much 
of that path length is "in COBOL" and how much is "in SQL". ABO won't fix 
the latter. It might fix the former.

I hazard as a "play number" that for a typical CICS transaction it might 
be 80% in SQL and 20% in COBOL (really application code).

So, in that case, ABO only attacks 20% of the txn's CPU. SQL tuning etc 
does the rest.

Now, I think we can get much more forensic about this putative 80/20 
split: In DB2 Accounting Trace we see Class 2 (and Class 7) CPU and also 
Class 1 CPU. Class 2 is "in SQL"; Class 1 includes that plus more CPU. So 
I think you have to subtract Class 2 CPU from any "region CPU" estimate 
you have. This ought to firm up the 80/20 which might easily be 50/50 or 
20/80. And then you have a reasonable estimate of the "ABO target". You 
then have to guess how much of that target CPU is saved with ABO. Or try 
it.

The above said "CICS". It might just as well have said "Batch", of course.



For what it's worth a little more "comp sci" explanation of the kinds of 
optimisations wouldn't go amiss. But the developers might consider that 
competitive information. I wouldn't know.

But I'm genuinely pleased ABO was released and think it will be really 
helpful for quite a few customers. The above is intended to be helpful in 
customers assessing the value.


Perhaps I threw away the chance with the above for a blog post (but I can 
still reuse it) and it has the makings of a presentation slide in my 
(accepted on the agenda in Munich) "He Picks On CICS".


Cheers,  Martin

Martin Packer,
zChampion, Principal Systems Investigator,
Worldwide Cloud & Systems Performance, IBM

+44-7802-245-584

email: [email protected]

Twitter / Facebook IDs: MartinPacker

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

Podcast Series (With Marna Walle): 
https://developer.ibm.com/tv/category/mpt/



From:   Jesse 1 Robinson <[email protected]>
To:     [email protected]
Date:   08/04/2016 20:51
Subject:        Re: Automatic Binary Optimizer (ABO)
Sent by:        IBM Mainframe Discussion List <[email protected]>



I never got any response to this question, so I thought I would contribute 
something. I got lots of help from the List writing my own COBOL program 
to make a preliminary foray into optimization ("COBOL Rookie Problem"). I 
got that program working but did not see much improvement with ABO. It 
does tons of I/O, which I'm told is not where ABO shines. So I tried the 
prime number calculator contributed by David Jousma. It also did not show 
much improvement, but ABO development pointed out that it actually does a 
lot of I/O by DISPLAYing each prime number found. 240,577 to be exact. So 
they replaced the individual DISPLAYs with a single DISPLAY at the end: 
"PRIME-COUNT 00240577". With that change, ABO showed a substantial 
improvement in the neighborhood of 27% reduction in CPU time vs. the 
vanilla version compiled with COBOL 4.2. 

The real question is how closely this program resembles a real life 
application program here. Doing only one I/O is pretty extreme, but the 
ratio of pure computation to I/O may be high enough to show improvement 
overall. I'm now waiting for availability of application programmers to 
pursue this further. 

.
.
.
J.O.Skip Robinson
Southern California Edison Company
Electric Dragon Team Paddler 
SHARE MVS Program Co-Manager
323-715-0595 Mobile
626-302-7535 Office
[email protected]

-----Original Message-----
From: IBM Mainframe Discussion List [mailto:[email protected]] On 
Behalf Of Jesse 1 Robinson
Sent: Sunday, March 20, 2016 11:10 AM
To: [email protected]
Subject: Re: Automatic Binary Optimizer (ABO)

Thought I would share an early experience with ABO. We are pursuing a 
'free trial' arranged by one of our IBM reps. (Yes, we still have them.) 
We are at z/OS 2.1, but the trial is also available for 2.2. I installed 
ABO its own CSI, figuring that we could move forward discreetly without 
impacting the mainstream maintenance process. However, one PTF has an 
ACTION HOLD that mandates installation of one z/OS and three LE PTFs. 
Quoted below. This is a bigger deal than I anticipated because it requires 
popping my head out of the prairie dog hole. These are pretty new PTFs. A 
very recent download of 'recommended' maintenance did not include them. 
Just a heads-up if you're thinking of pursuing the ABO trial. 

"++ HOLD(UI35271) SYS FMID(HALI110) REASON(ACTION) DATE(16042) 
   COMMENT 
    (Using this new feature on z/OS 2.2 requires z/OS APAR OA47689, 
     LE APARs PI52354, PI51546, and PI51802 be applied on the 
     system where the optimizer is used, as well as the system 
     where the optimized binaries are run. 
     Using this new feature on z/OS 2.1 requires z/OS APAR OA49419, 
     LE APARs PI55281, PI54804, and PI55010 be applied on the 
     system where the optimizer is used, as well as the system where 
     the optimized binaries are run.)." 

.
.
.
J.O.Skip Robinson
Southern California Edison Company
Electric Dragon Team Paddler
SHARE MVS Program Co-Manager
323-715-0595 Mobile
626-302-7535 Office
[email protected]

-----Original Message-----
From: IBM Mainframe Discussion List [mailto:[email protected]] On 
Behalf Of Jesse 1 Robinson
Sent: Thursday, March 10, 2016 11:18 AM
To: [email protected]
Subject: Automatic Binary Optimizer (ABO)

We are exploring ABO. This product alters the execution of pre-V5 COBOL 
modules in order to introduce-on the fly-some of the new efficiencies of 
V5 without having to recompile.

Has anyone on the List tried it out?

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] 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

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to