On Wed, 9 May 2007 14:24:41 -0400, Farley, Peter x23353 wrote: >OK, let me summarize the answers so far for the archives: > >1. (From several private responses) IBM has or used to have a performance >measurement tool called VLBPAA, possibly available from MKTTOOLS (see ><http://www.redbooks.ibm.com/redbooks/pdfs/sg242557.pdf>) that would measure >hit ratios and model (based on GTF data) what impact increasing #buffers >would have on hit ratios. > >2. Why bother measuring? What's the real cost of a few thousand buffers? >Can that be measured? It's only virtual memory. Virtual memory has a lost opportunity cost since its allocation removes the virtual address range(s) from the memory map. >3. Forget about BLSR, use SMB and a large buffer space instead. Oh, and >that also means use only SMS/VSAM Extended (if you're allowed to). > >I am surprised there are/were no basic statistics kept on such an obviously >performance-related tool as BLSR. But then, I guess if SMB replaces BLSR >then no one would bother going back to retrofit measurement tools into it, >either, regardless of whether it was hard or easy to do so. > >If it had been me, basic stats would have been built-in from day 1 for >things like actual count of buffers ever used for data and index, total BLSR >CPU time used, and maybe even hit counts and total searches. None of these >would have been hard to do and would have provided useful information to >users. And a report option for WTO of EOJ stats would have been >nice-to-have as well. And none of it would have required GTF. You could still write your own BLSR subsystem. Of course, your version of BLSR will have higher CPU cost since you intend to do all of that extra work and a lot of it will be inline (during GET/PUT/CLOSE). Maybe your BLSR could make the reporting be a user option so the user can decide? The statistics that I have always been most interested in having from BLSR vs. NSR have been the step CPU and the elapsed time. If the elapsed time improved 'enough' and the CPU was 'reasonable' then I could move on to other 'opportunities'. -- Tom Schmidt Madison, WI
---------------------------------------------------------------------- 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
