RE: Identification

["Tony Li" <Tony.Li@procket.com>]

Title: Message

 

I agree with most of what David said, except for this.  I interpret Jim's comment as asking

if the identifier is local to a particular location.  In other words, it acts as the low order bits

of the 'address'.  The issue with this is that it makes absolute identification a bit more

challenging.  For example, suppose that A, B, C, ... are locators and we have identifier Z.

 

A.Z and B.Z just happen to be the same host because it's multihomed, but then C.Z is another

host entirely.  What happens?  It means that hosts can no longer key on 'Z', and then have to

have other mechanisms so that they can determine that A.Z and B.Z are the same host, both

at an insecure and secure level.

 

If Z is instead global, then I believe that the identification problem is a bit simpler.  Consider the

role of looking up the TCB in your TCP implementation.  You just index by Z and ports and you're

done.  No futzing around trying to figure out if A.Z and B.Z are the right thing.

 

For this reason, I would tend to favor making identifiers global.  I think it's just simpler.

However, the other way probably CAN be made to work with enough effort.  In my mind, this

is one of the microarchitectural pieces that we don't need to argue about now.

 

Tony

 

 Can identifiers be within the context of Location?

Not sure what this means. Reachability to a given identifier depends on its locator, so perhaps I'd say yes.