[e2e] a means to an end
ali at sics.se
Fri Nov 7 09:30:20 PST 2008
Jon Crowcroft wrote:
> It is unusual kind of service for the internet community as it isnt an overlay,
> it has to work inside the IP layer
> as it is something end systems and routers need to share state/fate with,
> (we hit this before implicity with multicast, and couldn't ever face up to it properly)
This would preclude overlays, and in particular DHTs, because they're
app level (aside from ROFL, VRR and variants).
One of my favourite DHT papers is SkipNet by MS, which a lot of people
seem to have forgotten (not so cited anymore). They guarantee fate
sharing, even though it is an overlay.
I'll quickly recap the main idea for those who don't want to read it. It
is a DHT a'la Chord, but nodes on the ring are lexicographically sorted
(in reverse, e.g. se.sics at ali). Hence, nodes in the same AS would occupy
the same region of the ring. Hence, a partition would leave each
component of the partition fully functioning. SkipNet has some other
additional features, e.g. a) nodes don't need to be balanced on the
ring, and b) if you look for a piece of data within your AS you never
route outside of it, which is also good for transit costs. (I don't
remember the paper talking about AS:es and transit/peering links, but
it's essentially there).
So maybe we could use that?
More information about the end2end-interest