Send things to the list, not me.
On Mon, Dec 23, 2013 at 5:50 PM, Bill Broadley <[email protected]> wrote: > On 12/16/2013 12:01 AM, grarpamp wrote: >>> You may have a look of "I2P Bote" it is severless, encrypted mail >>> system, address is the public key, P2P based... nice tool. >> >> As in another post of mine, I'll be looking at that again. >> My first take was that it stores the messages in the DHT, >> which didn't seem scalable or reliable at all. I may be >> wrong as I read more later. > > I feel like I walking in halfway into a conversation, I'm guessing this > started on the cryptography list that I'm not on. > > Your DHT comment caught my attention though. What in particular about > DHTs don't seem scalable or reliable? > > Seems like DHTs are regularly in the 5-10M range and I don't see any > reason that DHTs couldn't be 10 times that. > > Any reasonable churn rate and reliability could be handled with > replication. The bit-torrent DHT for instance claims that 45% of users > that bootstrap from a central node are reachable 15 minutes later. So > typical setups involve 8 nodes per bin, and 20 bins. So every 15 > minutes you ping 160 hosts, only reach 45%, and do some work to > repopulate the missing slots. > > Given the simplicity of the bit-torrent DHT I think there's plenty of > room for improvement. Larger routing tables are obvious (at the cost of > more network bandwidth to track peers). > > The most promising idea for DHT improvements I've seen is to divide > peers into 3 latency groups. High, medium, and low. Much like L1 > cache, L2 cache, and main memory. That way common queries are very > fast, yet all queries still to find keys globally.
