Yes, I have been following this discussion...

First of all, Consoles does not keep any console information in the couple data 
sets. Secondly, the console attribute retention behavior has changed as a 
result of the Console Restructure.

Prior to the Console Restructure, when a console was activated for the first 
time, we fetched the attributes of the console either from PARMLIB or from 
OPERPARM depending on the type of console. If changes were made to any of 
the attributes of the console -- either while it was active or not -- those 
changes were retained and used whenever the console was reactivated -- as 
long as a simultaneously sysplex-wide IPL did not occur, in which case we 
fetched the attributes again from PARMLIB or OPERPARM. The reason this 
behavior occurred was because every system in the sysplex had a copy of all 
of the attributes for every console defined in the sysplex. Whenever an 
attribute was changed (by command), the change was propagated to every 
system in the sysplex. As long as there was one surviving system, it was able 
to pass the altered console attributes to any new system joining the sysplex. 
The only way to get Consoles to "forget" the altered console attributes was to 
make sure that -all- of the systems in the sysplex were dead before re-IPLing. 

Many customers were completely unaware of this behavior.

One of the problems we were attempting to solve with the Console 
Restructure was the amount of console attribute information that we were 
constantly passing from one system to another (under serialization) to keep 
everybody in-synch. We had situations where a system joining the sysplex 
would request a copy of the console information and before it could be 
completely received, the information had changed and had to be fetched 
again. We had some cases where the system never got out of that loop and 
the update storm then propagated to other systems in the sysplex, hanging 
the entire sysplex.

So as part of the Console Restructure, we made a concious decision to not 
have every system in the sysplex have a complete copy of all of the console 
information. (We had to back off of that somewhat due to some problems, but 
that was the intent). 

In z/OS R10, there is a new operations mode called SHARED mode that you 
can migrate to. In that mode, we only keep track of the attributes of active 
consoles. Once the console is deactivated, we forget its attributes, and when 
it is reactivated we refetch its attributes from PARMLIB or OPERPARM. All of 
the systems in the sysplex are still aware of all of the attributes of all of 
the 
consoles, but only the active ones. Also, we use a "lazy update" technique to 
propagate attribute changes from one system to another, but this is done 
using timestamp based serialization rather than the sysplex-wide ENQs that 
we used to use. In the past, the attributes had to be in-synch all around the 
sysplex because systems were making message routing decisions based on 
those attributes. We don't do that to the extent that we used to, so the lazy 
update is adequate. The primary reason we continue to propagate the 
attributes to all of the systems in the sysplex is so that DISPLAY CONSOLES 
will work.

W. Kevin Kelley  IBM POK Lab -- z/OS Core Technical Development

----------------------------------------------------------------------
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

Reply via email to