> It is a mistake to bind error control into the link layer
> protocol, and nowhere is that more clear to me than in radio.
MW> Hm, another example: Think of IP across an Ethernet. The
MW> physical layer is Manchester coding, the link layer consists
MW> of packets with a source and a destination address, a
MW> protocol identifier and a checksum - and CSMA/CD. Error
MW> corrention is not done at physical level. On dectection of a
MW> collision the link layer simply does Retries. Works
MW> perfectly. Not too much difference between this and doing
MW> hop2hop-retries on RF links.
But this supports my argument! Since Ethernet has extremely low loss, the
physical layer need impose very little overheard related to error handling.
(Manchester encoding is to facilitate clock recovery, not error handling.)
Link layer also imposes no error recovery procedure with Ethernet, and the
entire stack is best effort from IP down. A collision is very special case of
error, and link errors other than collisions are not retried.
-- Mike