On Wed, Aug 06, 2003 at 06:16:01PM +0200, Some Guy wrote: > I'm making the assumption that freenet nodes still > store trails of from where they recieved data for > specific keys. I'm not sure if NGRouting is going to > emliminate this table or not. The way I understand it > if a node gets a repeat request it answered before > that request goes to the node that answered before. > Please correct me if I'm wrong. > > Light Insertion: > Right now the way I understand it people reinsert > thier data on a regular basis to keep it alive inside > freenet. The whole things get copied down the routing > path from the inserter. > > Consider a light insert message, that follows this > path, but doesn't contain any of the data. Let's call > it an advertisement. Perhaps we could combine the > current insertion with a command to advertise. > > Ads could have a lower priority than other messages > and could be sent to nodes that complain about not > having anything to do. > Ads could be sent out with very large random HTLs > since thier sizes are minimal and they may be delayed > too. > Ads could be sent out more by nodes with low load to > nodes with low load.
We may do something like this, but not exactly like this ("inverse
passive requests") after 1.0.
>
>
> Ambience:
> Consider a random peice of data sitting on my node.
> None of my 100 neighbors know it's there. Now imagine
> if one of my neighbors, who specialises in that key's
> area, gets a request for it. What are the odds it's
> going to ask me? 1 percent! That's pretty poor.
No, most of the data stored on your node will be close to your
specialization. So there is a good chance that the neighbour will route
to you, if the datum is in your specialization area. Please read the
papers.
>
> In order for a node to be on a "hot trail" to the
> piece of data it must have downloaded and uploaded the
> data once. To make matters worse the trail is broken
> whenever one link dies.
>
> I'm sure there is a bunch of data lieing around
> freenet that is unreachable, even though requests for
> that data get within 1,2, maybe 3 hops of it.
>
> Solution: Nodes will send out "ambient" ads for things
> in their datastores with low HTLs.
>
>
> Both of these ideas are based on looking at the
> routing problem backwards. Instead of just focusing
> on how to get requests close to a target, we should
> also focus on getting links to the data close to where
> the requests will come from. Close your eyes and
> pretend you're a peice of data sitting somewhere on a
> freenet node and you want to get found.
>
--
Matthew J Toseland - [EMAIL PROTECTED]
Freenet Project Official Codemonkey - http://freenetproject.org/
ICTHUS - Nothing is impossible. Our Boss says so.
pgp00000.pgp
Description: PGP signature
