| Newsgroups |
gmane.comp.emulators.hercules390.general |
| Message-ID |
<[email protected]> |
On 6/22/2019 7:02 PM, kerravon86-/[email protected] [hercules-390] wrote:
> No application with a "quit" command
> should be producing non-deterministic
> results. None. At all. Not even applications
> that start with the letter "H".
Is that black or white magic???
When MVS issues an I/O, Hercules fields it, but some I/Os don't have
normal completion (e.g. outstanding 0A on a TP line), so Hercules can't
know when it's safe to close the thread.
On top of which Windows has an asynchronous write mode, so even if
Hercules flushes the buffers, there is no assurance that a disk write
completed.
If you look at MVS tape handling, you'll note that all non-data
commands are followed by a command-chained NOP (03). This ties up the
channel, preventing other access, until both commands complete; in the
days of selector channels, this meant that no other activity could take
place until the four minutes or so it took a full tape to rewind. And I
visited some government installation that put their DASD on the same
channel as their tapes! Since you already have modifications to MVS I/O,
you could add a NoOp at the end of the CCW chain, with the assurance
that the I/O completes. Also Windows should have a write-through option
that performs a write to completion.
In any case you should fix the shutdown commands to ensure all tasks are
complete, prior to issuing the P JES2,TERM / Z EOD / QUIESCE commands.
The QUIT after that should be clean. To make the JES2 shutdown easier,
ensure that all pseudo-devices default to DRAINED so you won't have any
funny hangs.
Gerhard Postpischil
Bradford, VT
---
This email has been checked for viruses by AVG.
https://www.avg.com