[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[freehaven-dev] Thoughts on general equilibrium for mix-acc paper



This post deals with the mix-acc paper draft.
http://freehaven.net/papers.html

Over the past few weeks, I mumbled a bit about general equilibrium analysis
for our network -- that is, when we make suggestions (create "contracts")
we need to consider the feedback mechanisms that would result.  Otherwise,
we would only be performing a partial equilibrium analysis, and ignoring
the effects of our actions.  There's quite a bit of literature in this
subject, both in economics and even in networking, often motivated by game
theoretic approaches.

It occurred to me, however, that we are already doing a rough form of
general equilibrium.  That is, the results of our "scorer reputation"
suggestion are getting fed back into the system, and future suggestions
reflect this process.

Time 0:  The network evolves a bit, nodes charge others with unreliability,
judges and/or scorers consider these charges, and assign some reputation to
nodes accordingly.

Time i:  Users look for paths, scorers relate these reputations (and
suggest paths accordingly.)

Time j (i<j):  Users continue to use the network, continuing to relate
unreliable nodes.  Thus, the judges/scorers continually revise their
estimates of node reputation based on this ongoing accountability mechanism.  

Time k (j<k):  When user's paths stop being reliable, they want to rotate
their paths for some other reason, or new users join the system, the
reputation measurement they see at time k has taken the feedback of use
since time i.  That is, if certain paths are getting too much traffic and
becoming less reliable (their bandwidth is hosed), this is reflected in
reputations at time k.  

Time l (k<l):  Users select other paths, the bandwidth of previously-full
nodes clears up, these nodes regain better reliability.

Time n:  This process iterates and the network stabilizes.  If we do this
correctly, I might submit that the network equilibria approaches
optimality.  Okay, that's too strong a conjecture, but you get my point.

All handwaving, of course.

Thoughts?
--mike


-----
"Not all those who wander are lost."                  mfreed@mit.edu