[Dave Hudson <dave@humbug.demon.co.uk> writes:]
>Any thoughts on using Ethernet CRC-checking instead of protocol-level
>checksums too? I've just been reading about some Sun Ethernet cards not
>doing things quite right and this getting through with a UDP
>implementation that didn't checksum (I don't know the full details
>though - it was a h/w problem apparently), but this would help with
>speed if we think it's OK. If not, any thoughts on a prefered checksum?
If we run directly on top of the Ether transport, then I'd trust to the
Ether CRC. IP's checksums were to catch end-to-end problems, across more
than one hop.
>The teeny server idea sounds like it'd be SLIPish, although PPP ought to
>be a reasonable bet for comms over a UART (I've not done PPP yet though,
>but it only looks like a couple of days of work). SLIPish sounds more
>in keeping with the objective of course (lighter weight altogether).
Yes, I'd avoid PPP with all of its baggage unless you need multi-protocol,
or authentication, or one of the other amenitities. SLIP's stateless encap
is otherwise much easier.
Andy
Received on Thu Jul 20 12:56:56 2000
This archive was generated by hypermail 2.1.8 : Thu Sep 22 2005 - 15:12:56 PDT