[Noisebridge-discuss] Consensus slides.
Jason Dusek
jason.dusek at gmail.com
Sat Sep 19 02:56:31 UTC 2009
2009/09/18 John Magolske <john at b79.net>:
> Thanks for the presentation & the link. [...snip...] One thing
> I was wondering was whether or not any of this might have
> applications in human to human interactions.
Actually, yes. People have no shared state (my brain is in my
head and yours is in yours) and we try to synchronize state
(come to and understanding) using speech and gestures
(messages sent on an unreliable network).
Some things to take away from the consensus problem when
considering people:
. Without a leader, you have `n^2` message cost (where `n` is
the number of people in a group). With a leader or small
number of leaders, this becomes `kn` message cost. For small
`n`, the hassle of appointing/handling leaders isn't worth
it; for large `n`, the win on message cost is worth a lot of
hassle with leaders. This offers an unusual perspective on
the role of leaders -- their job is to make communication
among the members more efficient. "To lead the people, walk
behind them."
. Message acknowledgement is a big deal. If you send a message
(communicate a thought) to a node (person) and don't wait
for acknowledgement before acting as though the message was
received, it's only a matter of time before you hit
inconsistency (misunderstanding). You have two ways to deal
with this:
. Recovery protocols -- an inconsistency, once detected, can
be smoothed over with a lot of messages once in awhile.
. Acknowledgement protocols -- to avert inconsistency,
always get acknowledgement before proceeding.
> when I first saw the title I thought it was going to suggest
> something to the Noisebridge consensus process :)
Although that's not what I intended, there is a natural
extension in that direction.
Consensus as we understand it won't scale. We see how the
membership adapted to this with the build meetings/debates. I
was one of many people who, after a little arguing, said that
they'd no longer participate in decision making but would show
up and help out.
--
Jason Dusek
More information about the Noisebridge-discuss
mailing list