[e2e] end2end-interest Digest, Vol 39, Issue 32

[Vadim Antonov]

On Mon, 21 May 2007, Randy Bush wrote:

> if there was 1/10 the engineering or science exchanged here as there is
> useless insults, this might be a worthwhile discussion.

Agreed.  However, the issue with Source Routing is in many respects
similar to the issue of Quality of Service features - there are many ways
to devise a network which supports the concept, but the problem is not in 
the engineering, the problem is that there is a fundamental philosophical 
disconnect between the design goals and the reality of what users and ISPs 
want.

The common utility-based arguments are based on the assumption that there
is indeed a way to engineer some quasi-economic system which makes
reasonable guesses about what particular communications are worth to the
users. Some proposed metrics of worth include bandwidth (resulting in
bandwidth-reservation schemes), ability to select a path, etc.

The reality, of course, is that we simply cannot know.  For example, a
specific 50-byte text message may have a much more worth to a particular
user than a specific multimegabit video stream.  So it is nearly useless
to try to optimize based on fixed metrics which cannot be directly
translated into explicit goals of actors.  These optimizations have
unknown (or, even worse, unknowable) benefit, while imposing real costs on
the specific actors.

The only way to learn if something makes economic sense is to see if 
people are actually using it and are willing to pay.

To the better, or to the worse, the standards (even the non-coercive 
standards, like Internet protocol specifications) have power to compel 
vendors to implement things even if nobody uses them - just for the sake 
of standards compliance and associated marketing advantage.

Since SR was around for decades, widely deployed, and well-known, and
still nobody uses it for anything but diagnostics, I think the question
whether it should remain in the standard should be re-examined.  Removing
it will relieve the router and host software vendors from the pressure to
implement it, thus resulting in cheaper and faster hardware, and will
reduce network maintenance costs (because of improved security and because
simpler equipment is more reliable).

This position, of course, runs against the claim that there is a need for
more academic research (and, by implication, more grants for research) in
this field. I think there isn't - there is a healthy natural evolution of
the technologies and business models in this marketplace, so any really
worthwhile idea will surely be implemented and tested by those who are
willing to take risks with their own money.

--vadim