How to identify bad performers?

baramesh · New User Joined: 20 May 2008 Posts: 25 Location: bangalore

I am new to performance tuning!, please help me understand, how do I figure out the jobs that need to be tuned from a huge set of jobs running on one LPAR of mainframe?. Similary, how do I make a list of CICS transactions that should need performance tuning ?

sergeyken · Posted: Sun Aug 15, 2021 12:51 am

I am new to nuclear power!, please help me understand, how do I figure out the nuclear reactor that need to be tuned from a huge set of devices running in one framework of a nuclear plant?. Similary, how do I make a list of heat exchangers that should need performance tuning ?

baramesh · New User Joined: 20 May 2008 Posts: 25 Location: bangalore

Rohit Umarjikar · Posted: Sun Aug 15, 2021 6:55 am

You need tools to do what you are asking , manually it’s not possible.
Understand what tools do you have installed at your site if you don’t have them then propose them to have one else you are going nowhere.

prino · Posted: Sun Aug 15, 2021 1:03 pm

Rohit Umarjikar · Posted: Sun Aug 15, 2021 7:41 pm

See if this helps.. found on internet
1. Go to SDSF;ST
2. Type "DA" in the "COMMAND INPUT"
3. Type "FIL TYPE EQ JOB" in the "COMMAND INPUT" to display only the batch jobs. If type is "STC", it is a started task. If type is "TSU", it is a TSO user job.
4. Type "SORT CPU% D" in the "COMMAND INPUT" to sort the jobs based on CPU% column in the descending order

Use ARR to rearrange columns. This shall give you highest cpu jobs and you might start from here for now.

Pete Wilson · Posted: Wed Aug 18, 2021 4:13 pm

It would be irresponsible for your employer to let you loose on the system without proper training or support.

You really need the support of someone with experience otherwise you WILL struggle, and at worst could make things a lot worse if you try and change things without fully understanding them.

Ask your site Sysprogs if you have RMF (or similar OEM product) installed. Learn how to use that properly and you will get to understand what to look for.

SDSF can help but RMF can process historical data as well which can be important if you have a transient issue.

You can also interrogate SMF data but that is not an easy task for anyone, let alone a beginner, you really have to know and understand what you're looking for.

Performance tuning has many levels to consider, which can be affected by the physical connections to devices and levels of load on channels, the amount of storage available in the processor, how the processor is split between LPAR's in terms of storage and engine weightings, WLM settings, cache size and settings in DASD controllers and the type of DASD, how many mirror copies of DASD and the channel lengths between them, internal dataset definitions particularly for VSAM, defined CICS buffers, Coupling Facilities, whether you have any dynamic buffering products such as Performance Essential, Block Sizes (BLKSIZE) of datasets, what products are running such as DFHSM and when it does it's major functions, when batch runs, what peak processing periods are. The list is endless, and one thing always affects another so changes must be understood in that context. You have to collaborate with many teams that cover all these different areas from Hardware to Storage to Sysprogs and Networks to vendors of hardware and software products.

hankoerlemans · New User Joined: 25 Jan 2018 Posts: 62 Location: Australia

The mad scientist approach can work.
Still using Cobol 4 ? What's the badest looking program or most critical ?
Maybe the Cobol 6 compiler can produce a better executable or IBM's ABO.
Try it and see.

In the good old days we used to start with SLA's.
If your batch SLA is being met comfortably then you hardly need to drop everything to go tune a mysterious "something".
Then if you're not then I believe you should do some critical path analysis to see where you are gonna get the best bang for your buck. Most sites use schedulers and it would be surprising if you couldn't get some salient information out of it.
Again with CICS. What's the SLA ?
Where is the focus ? Response time ? 10 people might need good RT to get through a defined workload, 100 doing the same maybe no.
These questions lend themselves to identifying tool sets - if you don't already have them.

Understand the information you are already collecting and what it means and how you can use it. RMF reporting ? MXG used to be popular.
Looked through IBM Redbooks ?

To do this properly you are now Sherlock and building a picture of the problem to arrive at a solution. Sure you can use a jack-hammer to split open an urgent problem but do yourself a favour and educate.
Performance management is more a science than a trade.

Too many folks on these forums are looking for a quick fix without doing the work. Some or us are less than polite at pointing at that out - so be careful.
As a former OS/390 Systems Programmer I used to tell people my greatest skill was Reading The Fun Manuals ! Goferit.

#disclaimer# I've worked with clever performance managers and picked up a bunch of notions but never worked as one.

hankoerlemans · New User Joined: 25 Jan 2018 Posts: 62 Location: Australia

Exactly Peter Wilson