Adam,
I certainly wouldn't be leading a crusade to reblock load libraries to save
space either. In most shops they make up a very small percentage of the
total space. I'm more interested in good practices when creating a loadlib.
I have always reblocked active load libraries, with some quantifiably good
results, and sometimes with no change whatsoever. Again it is not a crusade,
it is focused tuning. Starting out with 32760 as a Blocksize for loadlibs
means you don't have to go back and change it later.
There were many papers on this topic from the early days of XA fetch through
to the mid 90s when cache helped give fetch a bit of a boost. The following
is from a paper you may know:
"The effect of blocking, apart from device utilisation, can affect
Program Fetch performance. Small blocksizes will not conserve
storage, since 96K is always fixed for Program Fetch buffers, but
it may adversely affect performance. When a program is fetched,
one text record and up to 48 RLD/control records are read in each
I/O operation. PCI interrupt is used to add CCWs dynamically
and having larger text blocks allows more time for this activity."
The performance impact is less in cache controllers, but it still exists and
32760 is a reasonably easy way to ensure that fetch performance is optimal.
Ron
----------------------------------------------------------------------
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
