Re: [freehaven-dev] Micali's Exchange protocol

[Joseph Sokol-Margolis <seph@MIT.EDU>]

> I think this brings us back to square one with using a TTP,
> that's what we talked about a lot today...

My conclusion, and I think roger and david share it, was that using
any system comes down to that last problem the "how many acks is
enough" ttp solutions included. so since it boils down to a trust
thing why not eleminate the reciets and do it all in trust?

the buddy system should mostly track when a server is bad, and we can
have 3 buddies so there should be 2 in agreement....

In addition, when a server holding a buddy decides to check on the
other buddies, we want to avoid a) that server revealing it has the
data b) that server revealing it ever had the data and c) incorrect
buddy-death messages due to network outages.

buddy checks are anonymous and unsigned, so we don't have to worry
about them, but buddy-death messages are signed and they prove a given
person had the data, and you can assume they have the data for some
amount of time after an announcement. a risk: china sees there are
dissident tracks, china unplugs the network link to china, everyone in
china with a dissident track share get's shot)

if we have the buddy-death bit proxied through another servenet node
we accomplish a couple things. there's no proof a given node ever had
a piece of data. and if the machine can't reach a given node it can
assume a network outage. It picks a node randomly from it's node-list,
if the node happens to be down, eit there's no buddy death annoucement
this time. the machine will re-check later, or someone will. they
don't have to be immediate.

sorry this was so incoherent. it's oozing out my ears.

seph