Vernoy, If you have been following this thread you will see that track usage is not always better with larger blocksizes. I was corrected on this as well - for an LRECL of 7000 a BLKSIZE of 21000 wastes more of a track then BLKSIZE 7000.
As for performance on ESCON, the is a significant difference in performance of small blocks over large blocks when there is no IO chaining. However when comparing the difference of QSAM chaining 5x7000 blocks vs 5x21000 blocks the extra CCW processing and SSCH is hardly going to make you miss any batch deadlines. Before there were cache controllers every SSCH would add a half rotation to the elapsed time, but not so when cache controllers already have the requested data in cache. If you use SDB and get a BLKSIZE of 7000 then tripling the BUFNO will eliminate most of the extra SSCH, and there is only the small overhead of the extra CCW. Ipso facto roughly the same performance without wasting space. Ron > -----Original Message----- > From: IBM Mainframe Discussion List [mailto:[EMAIL PROTECTED] On > Behalf Of Vernooy, C.P. - SPLXM > Sent: Wednesday, 15 February 2006 11:57 PM > To: [email protected] > Subject: Re: DASD allocation guidelines > > "Ron and Jenny Hawkins" <[EMAIL PROTECTED]> wrote in message > news:<[EMAIL PROTECTED]>... > > Ron, > > > > With Sequential pre-fetch in all cache controllers nowadays I don't > think > > you will see much of a difference between BLKSIZE=7000 and BLKSIZE=21000, > > unless this is a very large file. If you want almost equivalent > performance > > then for SDB you can triple your buffers - DCB=BUFNO=15. Then you get > better > > space usage with the same performance. > > > > Ron, > > Better usage of what space? > > There are 2 other area's where BLKSIZE=7000/21000 will make a difference: > 1. Escon performance rapidly decreases with smaller blocksizes, I don't > know how much Ficon is sensitive to this, and 7000 is not really small, > but in general performance will benefit from larger blocksizes. > 2. Trackutilization is better with larger blocksizes. > > Kees. > > ---------------------------------------------------------------------- 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
