Dennis Heidner wrote:

> Gavin,
> I have no machines with check summing for the packets.
>
> With IPv6 the check sum field has been removed from the header
> packets.  The reason they were removed in the IETF RFC's is that the
> industry consensus was it was no longer necesary and not used by
> many of the routers/switches.

To insert a comment (and a correction to a prior reply of mine), the
actual case is that an IP frame carries a 32-bit CRC (covering the
entire frame) while the TCP payload has a 16-bit checksum.  I had them
backwards before.  The IP CRC is usually added by the NIC chipset but
the TCP checksum is still in the realm of software.  I *think* that IPv6
removes the IP checksum requirement as IP has no defined error recovery
mechanism anyway (OK, CRC check failed, now what?  It can only drop the
packet and log an FCS [frame checksum] error.  It is the responsibility
of upper layers to handle retransmission and error recovery.  So it is a
lot of extra circuitry to generate the CRC which is essentially
useless.  The "unused by many routers/switches" is because of the
complexity of generating the checksum (which varies with each hop as the
source/destination MACs change when routing).

The TCP checksum applies only to the TCP packet within the IP frame, and
is invariant through the life of the packet, even if fragmented and
reassembled.  It is true error detection and results in immediate action
(retransmission request), while a dropped packet due to IP CRC failure
is delayed until TCP realizes that it "missed" a packet.

Jeff Kell <[log in to unmask]>