At 1:52 PM -0800 1/21/99, John Border wrote:
> At 7:40 AM -0800 1/21/99, Jamshid Mahdavi wrote:
> > ...Receivers should, in
> > principle, be able to detect packets with failed checksum and send
> > back a CRACK indicating to the sender that a packet has been dropped,
> > but it is not a sign of congestion -- retransmit the packet but don't
> > reduce cwnd.
>
> We've looked at this concept from time to time over the years. One issue you
> have to deal with is that if the checksum fails, can you trust the fields in
> the packet you use to identify the sender of the packet.
Another issue is that the TCP/UDP/IP checksums are not very strong, and
if you turn off link-layer detection-and-discard of corrupted packets from
lossy links, you may significantly increase the risk that corrupted data
will be accepted (i.e., not detected) by receivers. The weakness of the
TCP/UDP/IP checksums is acceptable as long as stronger link-layer CRCs
are being used to detect and discard most transmission errors (so that
the TCP/UDP/IP checksums are mainly just defending against corruption
inside the routers and the host interfaces), but Jamshid's proposal implies
that those link-layer CRC checks would be turned off.
Steve
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 09:12:19 EST