On Thursday 13 November 2003 03:43 pm, jrandom wrote: > Assuming a magically secure and anonymous communication system, how would > you build a scalable content distrubution network? (or is it y'all's > [2nd pl poss] opinion that the anonymity attributes are inseperable from > freenet's scalability attributes?)
That all depends on how this communication system works. Does it work like some sort of limited depth directional broadcast system? That would be cool. > I'm not /just/ asking this for the... obvious reasons - there is a tie > in (& hopefully benefits to) freenet dev. Specifically, what is the > upper bound for CDN operations (latency & throughput)? I haven't followed > CDNs too closely (cdn is pretty close to the least of my concerns) - > if there are well known optimums for insanely heterogeneously (aka real > world) environments, a citeseer ref would do. > > Would NGR be optimal in an anonymity irrelevent system? What are the > CDN scalability properties of NGR specifically? Yes, and no. NGR under load should outpreform a strict DHT. It could be made better if we could know where each node was it terms of it's location on the network. The main gain is that we would not need pre-mix routing. But if it could also obscure which node actually returned the data, that would be nice, but it poses a problem. How do you know who to ask? If you ask everyone locally for the data, and then only the revelvent nodes process it, then how do you know that too many people aren't processing it? Or even that anyone is processing it? But it under ether system, the main performance hold up is the return path. If you had an underling anonymous network, you could return the data directly, but then how does one verify that the node that was supposed to return the data did? > To contribute at least a little to the discussion, my leanings towards > optimizing reachability (as opposed to throughput, bandwidth, or storage) > are currently pointing at a kademlia-style > (http://citeseer.nj.nec.com/529075.html) structured DHT. This would store > keys in $blockSize blocks using $hashFunction to balance the distribution > across the keyspace, with an associated $n bit hashcash calc to generate > node identity (& therefor specialization vs sybill) > > or is this totally off topic? > > (hi) > -jrandom _______________________________________________ Devl mailing list [EMAIL PROTECTED] http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl
