On Saturday, 01/14/2006 at 09:38 PST, "Harding, Mike" 
<[EMAIL PROTECTED]> wrote:
> Sorry for jumping into this so late, but I wouldn't rule  out shared 
segments 
> and would tend either to go that way or dataspaces.   Bill Stephens was 
> incorrect - you can define a segment as shared writable, I've  seen a 
couple of 
> apps built that way.  If you have more than one writer  you'll need a 
locking 
> mechanism (one I've seen used the first doubleword of the  segment as a 
> compare&swap lock) as well as a mechanism for overriding the  lock in 
case its 
> holder crashes.  As for Bill's other concern it's your  segment, you can 
define 
> it wherever you like and use diag64 services for  finding/attaching it.

A robust implementation of a shared R/W DCSS access method makes the 
locking and lock override mechanism somewhat difficult.

One of my early assignments at IBM was to work on the GCS recovery 
machine, whose primary purpose in life was to provide the needed lock 
release and data structure cleanup.  Not fun.

Alan Altmark
z/VM Development
IBM Endicott

Reply via email to