Exodus/C&W Depeering

Tue Mar 26 19:07:06 UTC 2002

| Won't this just increase the distance and AS count for Exodus outbound traffic,
| making Exodus hosting even less desirable?

Only in the minds of people who are lied to by Exodus's detractors.

I just spoke with the Invisible Hand of the Marketplace, and it
signed (in BSL, so the translation may be off) the following:

Usage-based billing: less performance -> less traffic -> lower bill.

C&W surely does not want to present a lower bill to you, and
thus may be motivated to figure out and correct what's going on.

Exodus's customers moving less traffic could take a (hypothetical!)
smaller bill to mean unpopular/stale content, or poor performance,
possibly near the end user, possibly near the on-Exodus source; if
the latter is suspected, distribute the content to elsewhere.

(If the former, tell all the politicos you can to stop protecting
local-loop monopolists!)

A (hypothetical!) larger bill could likewise be taken as
either increasingly popular content, or as better performance.
A _correlation_ with a change in Exodus's routing policy is
not the only possible cause -- someone elsewhere may have done
something to widen an intermediate bottleneck.

Remember that the AS path is only a trail of breadcrumbs used
to avoid route-announcement loops.  It is NOT a reliable indicator
of the forwarding path towards the destination.  An AS-to-AS "adjacency"
within an AS path *may* be completely meaningless.  You can draw
no conclusions about performance (in terms of loss, delay, and 
delay variance) by examining an AS path, although someone might
prove that for a given snapshot, from a given test location, there
are strong correlations.

Likewise, while from one particular vantage point, a "broader" peering
policy correlates with better performance (in terms of things which
actually affect goodput, namely loss/delay/delay variance), from other
vantage points, there is an opposite correlation.  This is not anything
remotely like a proof that strict routing and forwarding hierarchy
leads to better performance than relaxed routing and forwarding
hierarchy, with respect to how traffic is attracted towards a given
destination.   (Note that one can have strict routing hierarchy,
yet still observe a relaxed forwarding hierarchy: consider the
case where people do not next-hop-self at LAN-style exchange points).

Therefore, anyone predicting that the sky is falling, and big chunks
are gonna hit Exodus's customers and C&W (or predicting something opposite)
would be lying if he or she claimed to have a better view of the
future of this peering policy change than the average employee of
the psychic friends network.

	Sean.