Selfish routing

Mon Apr 28 16:39:24 UTC 2003

alex at yuriev.com wrote:
 >>alex at yuriev.com wrote:
 >>>If this is the last mile which is the problem, it will
 >>>*always* be a problem no matter if one talks about primary
 >>>or secondary path.
 >>
 >>Except when the last mile has an alternate (eg, a multihomed corporate
 >>site).
 >
 > Nope, then you either will get the congestion inside the corporate
 > network ( which you should not ) or the second path actually is the
 > best path.

OK; great.  And what mechanism were you thinking of to cause your flows 
towards that corporation to follow the second, better path?

For those following along in the general audience, imagine you have 
links to ISP A and ISP B, both of whom provide transit to corporation X, 
advertising prefix P.  Do you route P into A, or into B?  If the "last 
mile" link from A to X congests, how will you switch to the alternate? 
The main point here is that BGP contains no mechanism to detect and use 
higher performance, lower congestion paths.  It's not easy to make such 
choices promptly, correctly and stably.

Alex says that if we all just buy enough OC-192, the problem will go 
away; that's debatable, but immaterial, since the network is as we find 
it, not as we wish it to be.  I take the quote above as an 
acknowledgement that selection of path is important, and now the 
question is "how?".

On BGP's inherent choice quality (or lack thereof), I sent a pointer 
earlier to a single slide from a NANOG presentation.  That makes the 
point that BGP generates results quite close to what you'd expect from 
random routing (once you make it loop free).  I received requests for a 
pointer to a more readable presentation; for that, see the archive:
   <http://www.nanog.org/mtg-0206/ppt/mike1.ppt>

In a separate post, alex at yuriev.com wrote:
 > I do not have anti-vendor crusade. I have a crusade against
 > vendors who are selling ice to the eskimo, while molding the
 > concept of what ice is to fit their whitepapers.

I admit to being a vendor, and we do have whitepapers.  That may make us 
inherently evil in your eyes.  Fine.  But what about Roughgarden?  What 
about MIT's RON project?  What about the UWash Detour project?  Why are 
all these academic, non-commercial people pointing out that it's not a 
simple thing to build an optimal network, but it can be done using any 
of a variety of smart route selection schemes?

Are all these people just dupes of some grand pro-vendor conspiracy?  Or 
are you molding their work to fit your crusade?

Mike