Kacheong Poon wrote:
Kacheong Poon wrote:
This is part of the question 3. Is it OK not to "fully" support
the above setup? For example, by default an app in a system R
with only routable addresses cannot initiate communication,
meaning sending the first packet, to a LLA only system L. But if
the system L makes first contact, meaning sending the first packet,
to system R, then R knows how to talk to L as it knows where L is.
Is this limitation OK for the possible usage model?
I forgot to mention the failure mode of the above. Suppose
system R has two interfaces I1 and I2. And there are two LLA
only systems L1 and L2 which happen to pick the same LLA.
L1 is on the I1 network and L2 is on the I2 network. Then
the first system, L1 or L2, which makes contact with R wins.
It means that the other one will not be able to communicate
with R. Is this failure mode OK?
Sorry, didn't see this email earlier.
What I'm missing in all this is where lla would actually make the users
life any easier. We can come up with lots of cases where lla makes
things harder (either for the TCP/IP stack, the network applications, or
the users) but I haven't seen any positive statement yet.
I think it would be more productive to start with "we want to make X
easier" (could be for different values of X) and then see what is
needed. Perhaps lla is a critical piece of the solution or perhaps it is
a distraction; can't tell without starting from the problem we want to
solve.
BTW Have these top-down issues been discussed around MDNS/Bonjour? I
suspect that similar concerns around multihomed hosts appear there, but
I haven't though that through.
Erik
_______________________________________________
networking-discuss mailing list
[email protected]
