Re: two-phase send concerns

Larry Masinter <masinter@parc.xerox.com> wrote:
>I thought I started to understand this issue in the working group
>meeting, that the 'five seconds' was an arbitrary timeout, and that if
>the timeout were 'zero seconds' that this would also continue to work:
>senders should read if sending would block, and recievers should send
>abort messages as soon as they can, and NOT CLOSE THE CONNECTION if
>they care whether the sender actually gets the abort.
>
>Did I get this right?

I asked Dave Clark about the issue of abort and closing connections,
because I didn't understand the problem.  Dave (Clark) says a correct TCP
implementation should ensure than all bytes sent actually get to the
receiving end.  It's up to the receiver to actually look at them.

So, to return to Larry's question, I think the timeout could indeed be
reduced to zero.  However, apart from closing a connection correctly, I
don't think there's anything special a receiver (to be clear:  a server)
needs to do to ensure the sender (client) gets an abort message's bytes.
But the sender (client) does have to look for them!

Recall, though, that the point of the timeout was to give the server a
chance to avoid getting deluged by bytes.  So reducing the timeout to zero
is not necessarily wise.  But everything should indeed work.

Dave Kristol

Received on Tuesday, 5 December 1995 16:05:49 UTC