Just to clarify a few points from this thread:

The CFRM policy should specify INITSIZE for a structure only when that 
structure supports alter.  It's not meaningful otherwise.

The reason for the recommendation that SIZE should never be more than 1.5 - 2 
times INITSIZE is that when the CF initially allocates a structure, it must 
provide sufficient internal control objects to support the structure's eventual 
maximum size.  If SIZE is excessive relative to INITSIZE, it may be impossible 
to allocate the structure at size INITSIZE and still provide those internal 
control objects.  Allocation may fail entirely, or it may create a structure 
with so much of its storage consumed by internal controls that it provides 
insufficient objects (entries, elements, etc.) for application use.

Only DB2 GBP structures exploit user-managed duplexing.  DB2 (IRLM) lock 
structures and DB2 list structures use system-managed duplexing.

CFSizer and the SIZER utility are two different things, both available at the 
CFSizer web site (http://www.ibm.com/systems/support/z/cfsizer/).  The SIZER 
utility is the tool of choice for migrations of the type described by the 
original post.  It collects information about attributes and object counts for 
all currently allocated structures, and determines what size would be required 
to support the same attributes and counts in all CFs connected to the system 
where the utility is being run.  It is useful when you are satisfied that the 
currently allocated structures are adequately sized for the existing workload.  
If you wish to verify that a structure is adequately sized, or if you're 
introducing a new workload type or changing an existing workload, that's when 
you use CFSizer.   CFSizer requires you to provide input describing the 
application workload (peak ENQ count, message arrival rate and retention time, 
data base size, etc., specific to the application) and returns a size estimate 
based on that workload description.  As noted in a previous append, it 
deliberately produces generous size estimates, because an undersized structure 
can cause serious problems while it's practically impossible to go wrong by 
moderately over-sizing a structure (assuming you don't exceed the CF's 
available storage).

Bill Neiman
IBM Parallel Sysplex development

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to lists...@listserv.ua.edu with the message: INFO IBM-MAIN

Reply via email to