[e2e] What should e2e protocols know about lower layers?

[Vernon Schryver]

> From: Ted Faber <faber at ISI.EDU>

> ...
> And to beat the horse one more time, the checksum is a bad choice for an
> example of a protocol feature to turn off in local operation anyway.  

Yes, turning off the TCP checksum for "local" connections is bad idea,
but not as bad as turning off slow start.  (maybe using a cached initial
congestion window is not turning off slow start)

My point in recalling checksum debate was the hope of using it to
short-circuit some of the slow-start debate, not to reopen the checksum
debate.

>                                                                       I
> think you've done work (correct me if I'm wrong) that shows that as long
> as you've got one copy in the protocol processing path (and most hosts
> do), checksum computation can be rolled into the memory copy for free on
> modern CPUs - hide it in a delay slot.  In that environment,  the choice
> wrt to checksum is cut and dried - use it, it can only help you. 

When the insistent proposal to turn off TCP checksums was made, there
were available commercial systems from more than one UNIX vendor that
used various techniques to make the cost of the TCP and UDP checksum
0 nanoseconds/byte.  (yes, of course the costs of the checksum in
those systems in other measures such as instruction cache size, code,
money, gates, and complexity are not all 0.)


Vernon Schryver    vjs at rhyolite.com