David,

My understanding is that a generous BUFSP will improve performance when you
open a KSDS for sequential access, but for direct the NSR algorithm gives
you strings+1 and all the remaining BUFSP is used for Index buffers (some of
which are filled but never looked at). The "+1" is there for CI splits, and
the strings just process one CI at a time. I don't recall if DFHSM uses more
than one string.

I'm a great believer in using large CISZ to mitigate the impact of CA split,
and this applies to DFHSM. It takes less time to move a CYL of 18K CI than
it does to move a CYL of 4K CI one at a time during a CA split.

Typical NSR is just one string, therefore you usually get 2 data buffers and
the rest are index. I am not aware that HSM uses skip-seq access and it
probably can't take advantage of SIS. My recollection is that the BUFSP
allocation is decided at OPEN and does not change dynamically.

My thinking is that DFHSM gets more performance advantage from explicitly
specifying BUFNI to be slightly more than Strings+Index-Set and
BUFND=strings+1.

A long time since I was near this, and I'm happy to be corrected.

Ron


> -----Original Message-----
> From: IBM Mainframe Discussion List [mailto:[EMAIL PROTECTED] On
> Behalf Of O'Brien, David W. (NIH/CIT) [C]
> Sent: Thursday, 8 February 2007 10:52 PM
> To: [email protected]
> Subject: Re: FIXCDS to delete records from mcds
> 
> Ron,
> 
>   No, I did not mean to imply that BUFSP speeds split processing. A
> generous bufsp provides better overall performance thus masking whatever
> hits you may be taking from split processing. Not that I think splits are
> going to be noticeable in today's environment. I have seen sysprogs take
> the default here, possibly because the HSM IVP define lacks a bufsp
> definition. The starter job for HSM has a definition of
> BUFFERSPACE(530432).
> 
> I thought string+1 for data was the formula for CICS non-LSR files. Isn't
> the default for Direct Open 2 data and the rest Index?
> 

----------------------------------------------------------------------
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