[e2e] Overly Overlay; Peer to peer is commonplace

Jon Crowcroft Jon.Crowcroft at cl.cam.ac.uk
Sun Dec 23 04:24:19 PST 2001

so i've been reviewing lots of stuff on p2p and overlays (as many have
- probably partly prompted by the excellent CSTB report, and partly
just coz its cool work!)

anyhow, i keptbeing struck by rampant deja view - then i realized

IP is an overlay network - politically as well as technically, IP
spread its growth parasitcally over the underlying telecom
infrastructure, and newer overlays are doing the same - its a healthy
thing...we should think of it as the fight against entropyin the waist
of the hourglass - as soon as an overlay stops being experimental, it
evolves into the service, and moves downwards into the infrastructure
(IP on Glass..., maybe gnutella, CAM, Chord on photons:-)

But also, the peer to peer stuff was ringing lots of bells....not xmas
or front doorbells - so having implemented IP Routers twice (as well as
a 802.11 spanning tree system) i recall - ip routers are p2p - the way
you code a router (not enough space in this margin ...) is the right
way to do p2p...the system is tolerant of failed peers, is always i na
state of converging towards corerectness, without ever actually
reaching it, etc etc....

the big problem with ip (e.g. ospf, bgp etc) is that we have very very
few public domain high quality implementations (as compared to end
system code, where we have at least 2 [bsd and linux) - this is a
problem - the forwarding code in bsd and linux is fine, but we do not
have an example of an ioperational OSPF or BGP as far as i know that
is in really good shape....

so  thats why all these other p2p efforts are so useful, but it would
be nice to pull out the scaling lessons from (say) cisco's deployed
base of routing algorithms - there's a lot of things these systems do
to be robust in the face of broken peers, that the p2p end sys people
probably havnt encoutered yet (especially since they have no
real heterogeneity of implementations  yet etc)


happy winter solstice one and all!



More information about the end2end-interest mailing list