On Sat, 19 Sep 1998, Mitchell D. Miller wrote:

> Can I say you're both right?  Although TCP is a session layer protocol
- snip -
> provide, timeouts can occur if the tolerances are set too tightly.
> Unfortunately, many (most??) TCP/IP stacks don't allow you to adjust
> those tolerances.

Timeouts, even unadjusted ones, are not to be swept under the carpet by
the protocol implementation and/or the client, int the hope that no-one
will notice (ha, ha, ha - no-one notices 1-2 packets missing, they are a
low percentage of the total traffic, right ?  Now, where would someone be
who thinks that way, let's see...).

BTW, the last fragment has a flag to say that it is the last fragment. So
a timeout is NOT an excuse to consider the transfer valid EVEN if it is
ignored.

Peter