VSAM Performance Tuning..

Hariharan Ramachandran · Posted: Sun Dec 05, 2010 8:13 pm

Hi

I am trying to increase the performance of an VSAM KSDS dataset(Sequential processing). Currently the dataset does not support Extended Addressability. Now my question is..

1. Shall i use AMP=BUFNI,BUFND parmeter in the jcl or shall i use SMB technique(Ill have to make the dataset EA and then use AMP) ??

2. What are the advantages and disadvantages of the above two technique??

3. If anybody has experienced in using the above techniques please suggest me the best one ???

Note - I have already referred the REDBOOK Demystified.

Let me know if more informations are required.

Thanks in advance

#ari

PeterHolland · Posted: Sun Dec 05, 2010 8:26 pm

double post ?

ibmmainframes.com/viewtopic.php?t=52505

Once i had a utility to compute BUFNI, BUFND values. A pity i lost that.
For sequential access you define a larger BUFND than BUFNI, for direct
access the other way around.
You can always try the trial and error routine.

Maybe the next link will help :

www.mainstar.com/pdf/000-0513_CAT_NSR-Buffering.pdf

or this one :

www2.sas.com/proceedings/sugi24/Sysarch/p302-24.pdf

Hariharan Ramachandran · Posted: Sun Dec 05, 2010 8:53 pm

Thanks peter, i have already gone thru the PDF's now let me make my requirement much clear.

I have a dataset DATASET1 which is not in extended format. This is a master dataset and its been used at various PROC's. Depending on the cobol code for that particular step its access mode may be Direct or Sequential.

Lets consider my cobol has sequential access to DATASET1 which produces huge number of EXCP count and more elapsed time. If i wana increase its performance then which solution shall i go for ??

Soln1 - Ill have to define the DATASET1 as Extended format and use the SMB technique to increase performance

Soln2 - I could code the AMP=BUFNI,BUFND in the ddame of the DATASET1 to increase performance.

now please help me in deciding which solution i shall go for?? Which would be best and cost effective?? will the preferred solution will be acceptable??

If soln1 is best den how does it advantage over soln2 and vice versa ??

Thanks,
#ari

Pete Wilson · Posted: Mon Dec 06, 2010 2:20 pm

To reduce EXCP's then increasing the CISIZE and AMP='BUFNO=xx' values would help, regardless of whether it is Extend Format or not. The CISISZE should still be calculated to fit the MAXLRECL as evenly as possible, and FREESPACE (if any) should also be calculated to be sufficient to contain at least one maximum sized record otherwise it can end up as dead space.

You should really be reading the IBM's VSAM Demystified or Ron Fergusons VSAM File Structures to understand VSAM properly.

expat · Posted: Mon Dec 06, 2010 3:26 pm

Pete, The addition of buffers can also be one of the worst performance degraders if the file is access as skip sequential.

At the switch of access method all buffers are cleared and reloaded, so the more buffers you have the more resource it uses just to perform the refreshes.

PeterHolland · Posted: Mon Dec 06, 2010 4:34 pm

Expat,

with all respect i dont agree with that. CICS uses a lot of skip sequential
access. Beside that why doing skip sequential mixed with random and or totaly sequental? Skip sequential slows down when records are far apart in a large dataset.

expat · Posted: Mon Dec 06, 2010 4:41 pm

Peter, I agree, so was merely alerting the OP to the possible effects of making changes without fully understanding the consequences.

So give me a break mate, this is the first technical response I've given in ages, rather than the RTFM line of replies for the threads that deserved them. It was actually a little pleasing to be able to reply like that. Go on, go on, ruin my day, moan moan whinge whine, it's all you fault

PeterHolland · Posted: Mon Dec 06, 2010 4:47 pm

Expat,

i started with "with all respect", expecting you to be a little slow after all
that RTFM stuff.
But im sure you will be recuperated within no time.

Still with a lot of regards,

yours truly.

expat · Posted: Mon Dec 06, 2010 5:06 pm

Peter,

Just had a crap weekend, and today so far wasn't at all any better.

Get me a job over there and I'll buy you a beer or five

PeterHolland · Posted: Mon Dec 06, 2010 5:19 pm

Expat,

im really sorry your weekend was rotten. If i could i would have gotten you a job, but do you expect it to be better here?

But to have some beers doesnt need a reason, at least for me.

But (try to) cheer up.

Cheers,

Peter

Hariharan Ramachandran · Posted: Tue Dec 07, 2010 9:22 pm

Thanks

Hariharan Ramachandran · Posted: Fri Dec 10, 2010 11:58 am