On Wednesday 22 October 2003 11:49, Mike Stump wrote: > YMWaCI (Yet More Wind and Crasy Ideas): > > We should remove QR from the protocol, they are a waste of precious > bandwidth. Instead, queue inbound requests and answer them in last > in, first out order. Nodes sense overloading of other nodes by long > (to non-existent) service times. Also, the long service times will > push nodes that want faster response to other nodes, thus spreading > the load. Those requests that are answered, will tend to be answered > very fast, improving latency, which would be a good thing. Outbound > requests that are relayed go onto a very small outbound queue (built > 10-100 times a second), and in it we stuff very recent requests. > Those that actually hop, will hop with near 0 latency, thus all of the > successes will always have very low latencies through the node.
This is interesting. I don't know how well it would work if we weren't sending QR messages back, because then you would have to allot a certain time to wait for each hop before you assume it failed. However because you would vastly reduce the hop latency and the query responce time would vary a lot, it would have the nice side effect of thwarting most timing attacks. Although for the same reason it would require modifications to NGrouting code. IE: now the fail time is a known fixed constant. This also might encourage splitting requests or clients making the same request repeatedly to get it faster. Perhaps it would be better if we gave priority to nodes that don't make as many requests over those that make more. This way, a single node flooding you, or the network in general does not get their data as fast. > Also, we should tend to prefer sending answers of successful queries > of data requests from other nodes before sending answers to data > requests from our node, as odd as that sounds, so that we tend to push > load off our node to other nodes that obviously have too much free > time on their hand, if they can answer queries. The idea is to > increase the number of search sucesses a second a node can maintain, > which is good for all. Can you explain this better? _______________________________________________ Devl mailing list [EMAIL PROTECTED] http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl
