Home
Meeting Paul Harrisson from The Circle
Today we flew Paul Harrison up from Melbourne. He is the guy behind The Circle. It must have been pretty intimidating for him but we had a great couple of hours discussing technology - both where Sharman was going and his ideas behind The Circle.
The Circle is a great piece of work. Key points:
- Network scales very well. Each node needs to connect to its neighbors and log(N) other nodes across the circle. "The circle" here being the model for the network itself. If you do this sum you see that with 5 million people a peer would need to connect to 6 remote peers. Doubling the network to 10 million concurrents requires one peer to connect to 1 more peer. Scaling to 90 million only needs 8. It can grow huge. He had no worries about it growing and growing.
- Security concern. He explained that there was an implicit security issue with distributed hash table technology where someone could bring the network down very easily by disconnecting and connecting to the network. Need to understand this. He suggested that DHT technology needs to be massively encrypted to be protected and if it is not then someone can bring it down. So, current system based on the assumption that people are good. Kathey and I giggled...
- TC has user lookup as well as file lookup that works with a system of private and public keys. Seems to work very well. Nice touch: it creates a bezier curve based on the users hash which looks like a written signature - its great because it is humanly recognisable.
- Also a group system that allows collections of people to subscribe to groups. When a Message sent to all in group the peer literally sends out as many messages as their are people in the group. This is not good for groups of 100,000 so this probably would need a better solution involving passing messages on - sharing the load somehow.
- Trust system: "Gossip" which is modelled on Usenet floats people you trust to the top of the listing. Trust is also inherited so that if A trusts B and B trusts C then A inherits a bit of B's trust in C.
- Exhaustive searching is possible. When a user logs off then the index they are looking after is redistributed to other peers.
- Easy to add on other things other than files and people that are looked up.
Impressive and already containing so much...
Like what you're reading? Subscribe to my feed.
Today we flew Paul Harrison up from Melbourne. He is the guy behind The Circle. It must have been pretty intimidating for him but we had a great couple of hours discussing technology - both where Sharman was going and his ideas behind The Circle.
The Circle is a great piece of work. Key points:
- Network scales very well. Each node needs to connect to its neighbors and log(N) other nodes across the circle. "The circle" here being the model for the network itself. If you do this sum you see that with 5 million people a peer would need to connect to 6 remote peers. Doubling the network to 10 million concurrents requires one peer to connect to 1 more peer. Scaling to 90 million only needs 8. It can grow huge. He had no worries about it growing and growing.
- Security concern. He explained that there was an implicit security issue with distributed hash table technology where someone could bring the network down very easily by disconnecting and connecting to the network. Need to understand this. He suggested that DHT technology needs to be massively encrypted to be protected and if it is not then someone can bring it down. So, current system based on the assumption that people are good. Kathey and I giggled...
- TC has user lookup as well as file lookup that works with a system of private and public keys. Seems to work very well. Nice touch: it creates a bezier curve based on the users hash which looks like a written signature - its great because it is humanly recognisable.
- Also a group system that allows collections of people to subscribe to groups. When a Message sent to all in group the peer literally sends out as many messages as their are people in the group. This is not good for groups of 100,000 so this probably would need a better solution involving passing messages on - sharing the load somehow.
- Trust system: "Gossip" which is modelled on Usenet floats people you trust to the top of the listing. Trust is also inherited so that if A trusts B and B trusts C then A inherits a bit of B's trust in C.
- Exhaustive searching is possible. When a user logs off then the index they are looking after is redistributed to other peers.
- Easy to add on other things other than files and people that are looked up.
Impressive and already containing so much...
Like what you're reading? Subscribe to my feed.
Trackback Pings
TrackBack URL for this entry:
http://www.morle.net/mt/mt-tb.cgi/20
Comments
Post a comment
Movable Type 4.0 Everything here is my personal opinion.