On Sat, Nov 24, 2001 at 03:23:45AM +0100, Oskar Sandberg wrote:
> On Fri, Nov 23, 2001 at 09:08:11PM +0000, toad wrote:
> > ARKs would be extremely helpful in improving the efficiency of the network 
> > for
> > people with 24x7 nodes whose IP addresses change relatively frequently.
> > 
> > Is the following correct?:
> > 
> > An ARK is actually SSK@<node key>/<node address>, contents of which is the 
> > new
> > node address. Since we never need to look up a node address from just the 
> > node
> > key, this saves a lot of time messing with pseudo-updating.
> > When Fred's CP for a node falls sufficiently low (or it backs off?), it 
> > looks up
> > the above URL to try to get the new address of the node, if it succeeds, it
> > replaces the node address.
> > 
> > Oskar said in January that ARK = SSK@<pubkey>/<counter>. Is there any reason
> > to use a counter? If you use a counter, you have to store the counter for 
> > each
> > reference, and ideally include it when sending the pubkey/address pair 
> > through
> > FNP.
> 
> Well "<node address>" means very little in itself, since a node address
> can contain many physical addresses. What you could do is insert the
> update using the hash of the entire last reference (which is in FNP
> fieldset format) for the SSK name, but that does cost a little
> flexibility, since it means that a node can never attempt to skip
> forward a generation, and that you must make an ARK update every time
> the reference changes.
> 
> In theory at least this could be a problem, since you have situations
> where a node might talk to another using one physical address that
> continues to work while the reference is updated several times because
> of shifts in another (in practice nobody uses anything but tcp).
> 
> Honestly I don't care that much which way, but is certainly not a
> problem to keep a counter number in the reference (it's already
> implemented) and references are always transfered in the full (they are
> self signed) anyways. The references will need ARK specific fields
> anyways since there isn't a good way of getting away from keeping the
> encryption key for the data in there.
> 
> > Any other major problems preventing implementation of ARKs, apart from lack 
> > of
> > skilled-developer-time?
> 
> I don't want any major new changes implemented until the current
> codebase works decently. It's not too difficult for somebody who
> understands Fred's internals (dump a task on the Ticker, write your own 
> FeedbackToken and move into the request states) though.
Most permanent nodes actually run on DSL/cable modems. Not having ARKs
means they have to reattach to the network regularly (unless they set up
dyndns's, which is the exception, not the rule). Doesn't this have a
negative effect on routing? You mean, in terms of direct bugs in the 
codebase, not in terms of network upfuckage, then?

_______________________________________________
Devl mailing list
Devl at freenetproject.org
http://lists.freenetproject.org/mailman/listinfo/devl

Reply via email to