He might be taken more seriously if he took his work less seriously. It's not obvious that it will be useful, but he's clearly mining a beautiful and fun problem space.
I’ve met him a few times at small conferences and my ex used to babysit for him. I think he’s quite playful actually, in a very focused intellectual sort of way. He often goes extempore and makes contemporary jokes out of his presentations. This post too is sort of him testing himself on an unrelated topic, and the number of otherwise distinguished academics doing that is nearly zero (e.g. Roughgarden). But of course he has a very strong desire that the universe is actually just a manifestation of simple local rules (that would offer the best memory performance for the parallel processors in the simulation hypothesis, right?), and that all of mathematics can be replaced by the same.
The consensus problems is like a space in wich each "cell" have a different distance from other cells and this translate in different clock and differrent perception of time for each cell. And the cells can have random locations in space each time and jump in space all the time.. Modeling consensus as celular automata is interesting but maybe is just too abstract to get performant algoritms...
One insight is that consensus in a distributed system is that the most important aspect is the propagation of information betwen nodes. Practically, it is enough if you have a protocol imposing some logical epochs and time (real clock) limits after which the nodes are not alowed to change their mind about the events of an epoch. When you get information about the changes proposed by a large majority of nodes ( and what they know about all other nodes) than any kind of total ordering you introduce will give you a sound consensus.
6 comments
[ 2.8 ms ] story [ 28.9 ms ] threadOne insight is that consensus in a distributed system is that the most important aspect is the propagation of information betwen nodes. Practically, it is enough if you have a protocol imposing some logical epochs and time (real clock) limits after which the nodes are not alowed to change their mind about the events of an epoch. When you get information about the changes proposed by a large majority of nodes ( and what they know about all other nodes) than any kind of total ordering you introduce will give you a sound consensus.