Tobias,

I'm a great fan of RAID-0 datasets (dataset striping) and the benefits of
LSR buffering.

They are two very different things though, and if you would like to
understand the benefit of each technique you may want to measure them
separately.

Data Set striping does two things. It flattens skew and increases
parallelism. The benefit Data set striping may go unnoticed unless you run
some background IO to create contention, or run multiple jobs reading the
same dataset.

For datasets converted from NSR to LSR I usually find that the major benefit
comes from the buffering of the Index Set for a KSDS with three or more
index levels. You may want to create some KSDS with one, two and three
levels and compare your results with and without LSR. If the dataset has the
VSAM defaults then the three level index will probably realize a 50-67%
reduction in IO because of LSR. However if NSR is specified with
BUFNI=INDEX-Set+1 the LSR performance benefit will be much smaller.

In one shop I used ACC-SRS to force BUFNI=8 into the JCL of every KSDS
allocation which cancelled most of the benefit expected from LSR buffering.

A typical symptom to look for is a KSDS where IO count to the Index is three
or more times the IO count to the Data. This is a sure sign of three level
index, or greater, that is reloading the Index Set and a good target for
SMB, BLSR, or some old fashioned Index buffers. It would definitely make for
a good test program for your experiments. It is likely that optimizing this
IO challenge will discount the benefit of dataset striping.

Ron

> -----Original Message-----
> From: IBM Mainframe Discussion List [mailto:[email protected]] On
Behalf Of
> Tobias Cafiero
> Sent: Wednesday, September 15, 2010 8:17 AM
> To: [email protected]
> Subject: Re: [IBM-MAIN] DFSMS and System Managed Buffering
> 
> Ron,
>   I was expecting results similiar to the ones I saw in VSAM Demystified.
> Instead of using Repro, I'll  write something to do my testing. We will be
> using
> Dataset Striping and I am interested in putting them together.
> 
> Tob
> 
> ----------------------------------------------------------------------
> For IBM-MAIN subscribe / signoff / archive access instructions,
> send email to [email protected] with the message: GET IBM-MAIN INFO
> Search the archives at http://bama.ua.edu/archives/ibm-main.html

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html

Reply via email to