Exodus/C&W Depeering

Wed Mar 27 00:31:52 UTC 2002

At 02:04 PM 3/26/2002 -0800, Sean M. Doran wrote:
 >
 >
 >| This isn't something I really care to make a big argument of, but my point
 >| was that for many ISPs, the path will go from:
 >|
 >|    SELF - EXODUS
 >|
 >| to:
 >|
 >|    SELF - OTHER BACKBONE - C&W
 >|
 >| for a net increase in average path length.
 >
 >Are we talking AS_Path attributes here?  If so, all this means
 >is that now we don't announce OTHER BACKBONE routes to C&W/EXODUS,
 >which we probably weren't doing anyway.

Actually, it also mean a reduction in the possible paths presented to my 
router for computation.  Some would say this is a good thing.  Me, I like 
having multiple choices / redundancy.  Better to have two ways to get to 
EXDS than one.  IMHO, of course.

 >Or are we talking forwarding paths, which are _different_ (and not
 >necessarily stable even in the presence of perfect AS_Path stability)?
 >If so, have we added any bottlenecks or sources of packet corruption?
 >If none have been added, a change in the number of links and routers
 >traversed is meaningless.

The *number* of routers and links is probably irrelevant.  However, given 
C&W's history and current connectivity to "OTHER BACKBONES", it is likely a 
bottleneck, or extra fiber miles, or some other "problem" will be 
introduced into the forwarding path causing extra latency, packet loss, 
etc. for some portion of traffic.

Question is, do these problems affect enough traffic for anyone at C&W (or 
their customers) to notice / care?  If not, then C&W probably has nothing 
to worry about.

BTW: I am interested in what portion of traffic does not get forwarded 
along the expected AS Path.  Does anyone have an idea?

 >(Note that it is possible that the number of links and routers DECREASES).

Is is also possible the air molecules in the room where I now sit will 
suddenly all congregate into a 1 square inch space in the corner over 
there.  Although I admit your possibility is *slightly* more likely.

 >	Sean.

-- 
TTFN,
patrick