Re: [decentralization] Is peak bandwidth a scarce resource?
- At 10:02 PM 1/30/01 -0500, Clay Shirky wrote:
> > [...how to allocate overloaded local bandwidth...]This is a decent analysis of the situation at a macro level within a
>That having been said, I think there are three ways to allocate
>bandwidth during a /. effect; call the three Wait, Share, and Pay.
network system, but these are not the options available locally. There is
no difference between "wait" and "share" because they are in fact the same
solution for all but two P2P systems, Mojo Nation and the EDonkey 2000
Gnutella client (the only ones which allow for parallel downloads of
files.) In all other cases the local agent cannot get around the fact that
the data it is trying to share has not yet been replicated, because if it
had been you would have been directed elsewhere at the start.
Don't confuse the local decision-making process with system-wide
distribution and load balancing. If I and thirty others connect to your
host to try to download a MP3 file via Napster then all thirty-one of us
are going to share the bandwidth (the "wait" solution) and not until one of
us has completed the download and reported back to the central servers can
others who might have been directed to your PC be sent elsewhere.
At the local level, "pay" is not an option because no one is actually
implementing it (not even us, I do wish you would base your analysis of our
effort on more than a cursory skimming of an out of date FAQ...) By having
the tools available we can implement this sort of a system quickly using a
PMP model to allow agents to bump themselves to the front of the line, but
until it is necessary there are easier ways to get around the problem.