If you can get a hold of it, I recommend a paper from CMG 2000: "VSAM and Non-VSAM Buffers - How Many are Enough". I know I have a printed copy, but I'm not sure if I have a soft copy to send. Obviously the paper is slightly dated at this point, but is still a great piece of work.
Also... The following is from something I wrote on search30.com: Buffering is an area that can be very confusing and I have seen many different ROTs and recommendations over the years. Here is a statement from a Redbook called System/390 MVS Parallel Sysplex Batch Performance: For non-striped data sets, QSAM defaults to five buffers. When dealing with striped data sets, the default changes. The default number of buffers for QSAM for striped data sets is two times the number of blocks per slice. This allows QSAM to automatically adjust for changing block sizes and changing numbers of stripes. The documentation goes on to say this: ...QSAM will not transfer more than five slices worth of data per I/O packet. Using these rules, it can be seen that the new default number of buffers (two slices worth) will allow QSAM to do double buffering by default. This new default means that a data set processed with QSAM with even only one stripe should get a performance benefit because of the double buffering. Previously, over 240K worth of buffers was required before any double buffering was done for QSAM. The recommendation for QSAM is to use the default buffering as this will provide acceptable performance and does not require any changes as the number of stripes changes. Larger BUFNOs should only be specified for critical jobs that are still I/O bound. Note: It is very important that an intelligent number of buffers is used. If too few buffers are used, the performance can actually be worse than for a non-striped data set. Therefore, when converting a data set to a striped data set, be sure to check all jobs that use that data set for any buffering options used. So, to answer your question, the general guideline is to use the default buffering. If you do specify your own BUFNO, at least 2 slices worth of buffers must be specified to achieve double buffering so a complete slice can be read in parallel. For more information, please see the manual I referenced above: System/390 MVS Parallel Sysplex Batch Performance (SG24-2557-00) This can be found and downloaded by searching the IBM Redbook site - http://www.redbooks.ibm.com Also see DFSMS/MVS Using Data Sets - http://publibz.boulder.ibm.com/cgi- bin/bookmgr_OS390/BOOKS/DGT1D411/ -- Mark Zelden Sr. Software and Systems Architect - z/OS Team Lead Zurich North America and Farmers Insurance Group mailto: [EMAIL PROTECTED] Systems Programming expert at http://Search390.com/ateExperts/ Mark's MVS Utilities: http://home.flash.net/~mzelden/mvsutil.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
