----- Original Message -----
From: "Harik" <[email protected]>

> I was thinking about this for a few days and there's a way to implement
> 'bundles' without massively duplicating data: Parse the metadata, and
> see if any of the listed keys are available in the local store.  If so,
> return them along with the requested data.  For each hop on the
> return trip, add to it.   Using appropriate size/count limits would
> implement prefetch without 'hiding' CHKs inside zip files.

That won't work for a few reasons.

The first reason is that all data on freenet is end-to-end encrypted,
meaning that the nodes inbetween the inserter and the requestor can't see
any of the data (and although there is metadata associated with the
encrypted data, it is different metadata than what you are talking about)
nor the metadata. Therefore, they can't parse the metadata they don't have
the decrypt key for.

Also, since the route a node follows to find the data should hone in
specifically on that data, in a well routing system, it should be rare to
find other thinks linked to from the metadata just lying on a node it passed
through.

Hmm. If you added the additional keys which came with the metadata to the
routing tables of all the nodes after it (by this I mean, you parsed key
chk at aaabbcc and it's not being supplied with the metadata for the end user)
you could extend the path lengths and hurt the routing tables, so doing this
would be a bad idea [note: this part doesn't hurt the argument, just cuts
off bringing it up in discussion :)]

> Similar to browser prefetch hints, but done at the FNP level.
>
> Also nicely obscures the size of the requested data, because some
> random amount of 'extra' data is being returned as well.
>
> (Possibly strip off the individual file padding then pad the bundle
> out?)

That would be an expensive process redoing a chk like that on the fly. It
does have a positive effect though. Still, the idea probably won't work.

> --Dan

-Mathew


_______________________________________________
devl mailing list
devl at freenetproject.org
http://hawk.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to