RE: Simple dual homing experiment

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

Perfectly reasonable.

I should point out that turning up an additional link today
breaks connections because the host isn't capable of trying
a different address.  You could, in theory, write a non-standard
TCP that had the right host behavior, even with IPv4 today.  But
it doesn't exist, AFAIK.

Now, does this test actually distinguish anything between any of
the proposals that have been discussed?

Tony


|    -----Original Message-----
|    From: Christian Huitema [mailto:huitema@windows.microsoft.com] 
|    Sent: Thursday, March 20, 2003 11:49 AM
|    To: Tony Li; kurtis@kurtis.pp.se
|    Cc: multi6@ops.ietf.org
|    Subject: RE: Simple dual homing experiment
|    
|    
|    I agree that exit selection (3.5) is more of a "nice to 
|    have" -- the real requirement is, try to achieve some load 
|    sharing so we do not get a gross misuse of the resource.
|     
|    The reason for "should not break" rather than "must not 
|    break" is that in a similar set up today (e.g. main 
|    connection backed up by a DSL line), turning on backup 
|    actually breaks connections. Also, I am not confident that 
|    we can make sure that no connection will ever break, so 
|    "must" may well be slightly out of the reach of engineering.
|     
|    By the way, your general observation is correct: David an 
|    I believe that most of the solution is actually fairly 
|    easy to engineer.
|     
|     
|    
|    ________________________________
|    
|    From: Tony Li [mailto:Tony.Li@procket.com]
|    Sent: Wed 3/19/2003 8:22 PM
|    To: Christian Huitema; kurtis@kurtis.pp.se
|    Cc: multi6@ops.ietf.org
|    Subject: RE: Simple dual homing experiment
|    
|    
|    
|    
|    |    David and I wrote up a "simple dual homing" proposal, 
|    |    based on previous drafts and private conversations. This 
|    |    may be one of several ways of going forward in multi6. 
|    
|    
|    Christian, 
|    
|    First, I don't believe that 3.5 is a real requirement.  It is 
|    not a problem that is solved today in any technology short of 
|    full blown traffic engineering.  While I agree that this would 
|    be valuable, I don't see how this really cuts it as a requirement. 
|    
|    Of your other requirements, it would seem that they would all 
|    be addressed as long as: 
|    
|            1) routers which don't have a working exit don't advertise 
|             default 
|            2) hosts rotate their source (and destination) addresses 
|               among the possible set 
|    
|    Further, I would argue that your requirement for 3.4 isn't strong 
|    enough.  That the connection should NOT break. 
|    
|    Comments? 
|    
|    Tony 
|    
|