The structure size has always been overly conservative, especially given that 
signals are usually in and out, which means the structure is mostly empty.  
However, the target systems can become unresponsive, which means signals can 
pile up in the structure.  The CFSIZER goal was to ensure that any given pair 
of systems would be able to continue sending signals even if all the other 
systems in the sysplex seized up and stopped pulling signals from the 
structure.  Thus every pair of systems would be able to obtain its fair share 
of list entries in the structure, thus avoiding sympathy sickness.

To the extent that the systems are responsive, then you do have a lot of excess 
capacity.  On the other hand, that also gives the structure some leeway to deal 
with cases where "large" signals get sent via a signal structure sized for 
"small" signals.

For a sysplex consisting of systems running z/OS 2.4 or later where every 
system has the XCF function switch XTCSIZE enabled, any size signal can go via 
any structure.  For what I believe to be a typical sysplex -- one where those 
956 byte signals are 90% or more of the signal traffic -- I tell people to size 
the structures for a classlen of 12220 (12K-68) as that should make the 
structure big enough to satisfy the above principals on average.  Of course, 
that will make you even less happy with the structure size.

Mark A Brooks
z/OS Sysplex Development

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to