On Wed, Jan 20, 2010 at 11:00 AM, Matthew Toseland <t...@amphibian.dyndns.org> wrote:
> Well, in terms of bloom filter sharing, we will never want to transfer such > data unless we have seen a node regularly. It will take more than a few hours > to send the data, especially if we are using spare bytes not used up by > requests to do it. It is very costly and a long term commitment. It works > fine on darknet, but on opennet it is pointless unless we are fairly sure > that we will see the node again soon. Which means it must have been connected > for many hours over the last month or so. And yes it is possible that bloom > filter sharing on opennet is too costly; there are lots of other things we > want to do to optimise data retrievability, and in particular to improve > opennet announcement and reconnection and generally support for low uptime > nodes, before tackling bloom filter sharing. We've talked about partitioned bloom filters before, such that having a partial bloom filter is partially useful. Assuming we do that, there's a (conceptually) simple approach we could take on opennet: just transfer the filter using the bandwidth that would currently go to padding bytes. Nothing changes about how many bytes we send total or of regular traffic, so there's no harm even if we don't ever see the peer again. Peers that we see often enough will have enough of our bloom filter to be useful. (I agree that bloom filter sharing should come after other retention improvements.) Evan _______________________________________________ Devl mailing list Devl@freenetproject.org http://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
