NetDb lookup reduction and negative cache improvements

[zzz]

Copied from ​http://zzz.i2p/topics/1996 post 13:

Another theory I've just started thinking about is the number of lookups in the network. On my floodfills I'm seeing peaks of about 250K lookups an hour, or 65 a second, with about 75% failing.

I'm thinking of reducing the max floodfills-to-search in the router and some other tweaks.

But also thinking about bittorrent DHT as perhaps a big source of the lookups. Is the 8-hour expire time too long? Does Vuze do enough caching of hash→Destination? Is either Vuze or snark too aggressive about DHT lookups and refreshes? Do we need more negative (or positive) caching in the router, in I2CP, in snark, in Vuze? Either in time or in size of the caches.

There's also a guy looking in the other direction, he thinks the negative cache time in the router could be too long, and that it's causing problems for IRC reconnects. Also, the router's negative cache is pretty dumb, we just clear the whole thing every couple of minutes, not per-entry. But any longer caching would probably need to be done out in the clients.

Early times, just started research in this direction.

[zzz]

Initial changes in 2aac9dd756df2f2dc6dc6560310760ca418a9476 0.9.23-12

Increased floodfills starting about December 5 has also improved things. See zzz.i2p thread linked above.