There's some coverage of techniques which scale to the hosting of hundreds of thousands of processes per node across indefinitely many nodes, associated with the GPLed Java-based process hosting toolkit called Diet Agents... http://diet-agents.sourceforge.net
This is itself the foundation platform of choice for continuing EU research projects in scalable autonomous systems... http://www.cascadas-project.org/ ...and some of our own work in scalable hosting... http://cefn.com/blog/btok.html The main emphasis here is on handling resource contention and providing core operations only where the resource load is systematically independent of the number of resources in the system. The load of these core operations is independent of the number of processes and interprocess communication channels locally and independent of the number of hosts globally. There's a bit more in the tutorial... http://diet-agents.sourceforge.net/.rsrc/tutorial.html#diet-philosophy ...but of course the code is its own best description :) Cefn http://cefn.com/blog/ -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Jeff Rose Sent: Tuesday, July 10, 2007 2:18 PM To: theory and practice of decentralized computer networks Subject: [p2p-hackers] Super Jumbo Uber Massive Networks Yo, I read a good paper the other day talking about building optimally random networks (one of these [1,2] I don't remember which, but both are worth reading), and one of the ideas popped out as being obvious but seemingly lost in a lot of P2P literature. They call it elasticity, the ability for a network to grow indefinitely without diminishing in performance. (So the amount of work done by any node in the network can be bounded without respect to the size of the network.) Even in a lot of unstructured P2P systems there are fundamental issues with scaling up many of the algorithms people publish. The world currently has about 1.1 billion internet users. If trends continue many more will join and a huge portion of these users will want to run their own sites, servers, clients etc. If we want to make the future peer based, I think scalability is going to take on a new meaning. So, with that in mind I wanted to ask if people know of any cool and interesting ideas, papers, or systems that they think could scale up to massive sizes. (Lets say on the order of 100 million to a billion nodes.) I'd like to put together a collection of P2P work that really scales, and I'm more than happy to put it up in a public place for everyone view. -Jeff [1]: http://people.inf.ethz.ch/spyros/papers/Gossip-based%20Peer% 20Sampling.pdf (Journal paper) [2]: http://people.inf.ethz.ch/spyros/papers/Thesis-Voulgaris.pdf (Thesis, long but interesting) _______________________________________________ p2p-hackers mailing list [email protected] http://lists.zooko.com/mailman/listinfo/p2p-hackers _______________________________________________ p2p-hackers mailing list [email protected] http://lists.zooko.com/mailman/listinfo/p2p-hackers
