IBM Mainframe Discussion List <[email protected]> wrote on 12/08/2007 
10:45:02 AM:

> First, since the SRB is a unit of work that is independent of any 
> existing dispatchable unit in the Master, any normal error that 
> occurs within it has
> no impact on any other work in the address space.  If it abends, it 
> doesn't take a TCB with it.  (The SRB must still use appropriate 
recovery,
> though.) If the SRB schedules an IRB, the IRB abending can take an 
> unsuspecting TCB down.  Of course, the SRB could overlay storage in the
> Master and cause a systems crash, but anything running in Key 0, 
> Supervisor State in ANY address space can cause enough damage to
> bring the system down.  Running in the Master probably increases 
> this by only a very small percent.
> 
> Second, the blame really goes back to IBM's design of CADS.  By 
> their very nature, any CADS should have been made persistent and NOT
> tied to the life of the creating address space, i.e., it should have
> been maintained just like CSA/SQA and not automatically freed when the
> "owner" terminated.  So the only place to create a CADS that must 
> never go away is the Master.

  CADS were not part of the original data space design in 
SP3.1.0.  They were a quick add-on in SP3.1.3, and as such,
inherited the property that every data space is owned by some TCB
in an address space.  If you want to make sure that the 
CADS does not terminate, the address space must be 
non-CANCELable, non-FORCEable, and non-memtermable
(so you must turn on ASCBNOMT).  I agree that it would be
preferable to allow a value of zero for the TTOKEN on DSPSERV
CREATE and DELETE for SCOPE=COMMON data spaces (as always, feel
free to write a requirement).  In which case, internally, we 
have the CADS be owned by Master, associated with a TCB address 
of zero, so that no task termination could match and delete
delete the CADS.  Implemented this way, DSPSERVE CREATE
could probably do this directly under the workunit of the
DSPSERV CREATE, without needed to schedule an asynchronous
workunit in Master. 
 
 
Jim Mulder   z/OS System Test   IBM Corp.  Poughkeepsie,  NY

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