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

[Vernon Schryver]

The arguments for turning of TCP congestion avoidance and recovery
for "local" networks smell as bad as the arguments for turning off 
TCP checksums on the grounds that link layers CRC's are good enough.
In both cases, they're based on things that are sometimes but often not
true and that cannot be determined reliably.  Both have been based on the
bogus notion that the netmask has something useful to say about "localness."

"Local" is not the same as "netmask says use ARP" except for a trivial
and essnetially useless notion of "local".  That notion is useful only
for routing, not for speed or security or much of anything else, which
is just what you'd expect because the netmask is merely a routing notion.

The questions of what a link layer should tell an upper
layer or what makes a "local" TCP connection are intersting.


Vernon Schryver    vjs at rhyolite.com