--------
> Bora> If there is extra information available maybe it
> Bora> should be used.
>
> Precisely my question, better stated (thanks). How does one intend to
> use a bit in a connectionless oriented datagram to communicate meaningful
> congestion vs. link noise (for example) back to the sender of the datagram?
>
> I find this both interesting yet orthogonal to IP. Could someone please
> describe this 'concept' in protocol level details. What is causal to what,
> feeding back what information to whom?
>
There is plenty of analysis out there that states that with a physical link
with Bit Error Rate (BER) of 10^-7, TCP gets very limited throughput
(approximately 30 percent or less) due to the losses being interpreted as
congestion and premature clamping on the window. Hence I think it is WELL worth
the effort of making TCP/IP work over links with physical errors.
For 1500 byte packets 10^-7 translates to roughy 0.0012 packet error rate
which is much less than 1 percent but the throughput is still dismal.
I believe this is why we may be interested in getting some sort of information
about the corruption vs congestion loss and acting accordingly.
Again SCPS-TP protocol describes how this can be implemented and
there is a recent paper out from Randy Katz's group at Berkeley
on this issue. These may be suggested readings.
Bora
Bora Akyol
BBN Technologies
GTE Internetworking
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:30 EST