[e2e] Overly Overlay; Peer to peer is commonplace

[John Heidemann]

>
>In message <5.1.0.14.2.20020101103822.02b81908 at mail.reed.com>, "David P. Reed" 
>typed:
> 
> >>Wireless networks, especially densely scaled mobile wireless networks, do 
> >>not behave like "wires without wires" or "fibers without fibers".  Topology 
> >>is not naturally hierarchical in its interconnection, for example.  So 
> >>"hierarchically derived" topological addresses are just plain wrong.  More 
> >>relevant, though again as naive as GUID-based routing, is geotemporal routing
> 
>
>the similarity of the recent parallel work on
>smart, ad hoc, self organised, wireless network routing
>and
>smart ad hoc, self organised peer2peer systems, 
>has been remarked on a few times...

But the similarity shouldn't be exaggerated.

Christian H. just commented that many of the ad hoc protocols scale
O(n) or worse (where n is the number of nodes or even number of
flows).  This approach is quite reasonable for an interesting class of
ad hoc net problems, but it's quite a different constraint than the
general Internet where scale in numbers of nodes is central.

By contrast, many peer-to-peer systems that push scale operate on the
assumption that "all nodes are pretty close to each other".  For
example, FreeNet and Chord both basically hash content to nodes
largely irrespective of node's network location.  For nodes operating
in the Internet, this assumption is quite reasonable.  But I think one
would not be happy trying to apply this to ad hoc networks.

More similar to ad hoc networking are some of the peerish overlay
network work.  For example, the Resilent Overlay Network paper that
appeared at SOSP last year which showed reasonable performance
improvement due to link-state routing with frequent updates but
explicitly didn't try to scale past 10s of nodes.

   -John Heidemann