[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Source address selection insufficient?



> I think that this is getting more complex than necessary.
> It is true that routing is frequently asymmetric.
> However, that is a very different statement from saying that connectivity 
> is asymettric.  While the routing path among ISPs for a given address pair 
> (src, dst) may be different in the forward and reverse direction, it would 
> take a very strange situation for a pair to work one way, and not when 
> reversed.  It would take an even stranger situation for there to be no pair 
> that worked in both directions.

Things might be a bit subtle - perhaps we don't understand well
enough the interaction between routing, source address selection,
and ingress filtering yet.

A few observations:
1. If you ignore the effect of the source addresses and ingress filtering,
   all paths between A and B work in both directions in my example.
   Thus there is no need for unidirectional failures of links to get into
   this state.
2. Each link between a site and their ISPs can have
   packets pass in both directions in the example; there exist a <src,dst> 
   address pair which makes packets traverse a particular site/ISP link.
3. One perspective is that we get in trouble due to the assumption by the 
   transport protocols that the same address pair is used in both directions;
   the assumption seems counter to the loose notion of "address selection" -
   only one end of the communication can select address.
   (Another perspective is that ingress filtering doesn't fit.)

   
> As such, I think we ought to be able to assume that there exists (at any 
> given time) an address pair that is useable in both directions.  (The 
> packets using that pair may not take the same path forward and backward, 
> but will use the same ingress / egress points on each end.

I don't understand.
From the fact that communication is possible in both directions, both e2e and
over each of the site/ISP links, it doesn't follow that an adress pair exists
that can be used in both directions.
Can you clarify?

  Erik