At 11:15 AM 5/16/00 -0700, David Schnepper wrote:

>
>Jim --
>Events now use out-of-band if the stack implementation on both ends supports
>it. Otherwise it falls back to the separate socket approach.
>

What are the advantages of OOB over a separate socket?

>Events were (are?) the most trouble prone area of any IB port I can recall.
>

That is because you missed the fatal signal/thread bug on Apollo that
delayed V3 by many seasons. Turned out that if a signal arrived during
thread switch their kernel pretty much lost its mind.

But you are correct. Asynchrony is where almost every operating
system's really ughly warts hang out. The only exception was VMS,
which handles asynchronous operations reasonably gracefully (in other
words, the synchronous cases were just as grotesque as the asynchronous
ones).

For the record, I lifted the basic event count mechanism from Apollo
Domain, extending it with named events and network connections.


Jim Starkey