Hi Oskar,
On Fri, Jun 15, 2001 at 11:36:39PM +0200, Oskar Sandberg wrote:
> On Fri, Jun 15, 2001 at 01:38:46PM -0700, Ray Heasman wrote:
> > Hi Oskar,
> < >
> > > Maybe we need to accept the fact that our model has, thus far, failed
> > > completely and disastrously in pratice. Not to say we should give up:
> > > there are still a lot of unexplored options, but the current network is
> > > very obviously useless. All this beating of the dead horse is doing
> > > nobody any good...
> >
> > I agree. I honestly think that the current routing model can be made to
> > work. I also honestly think that the way Freenet caches data is what's
> > breaking the routing. I know I'm beating this drum regularly, but I've never
> > been able to stimulate much discussion on the subject, and I'm not going to
> > give it up until I feel I have...
> >
> > It would be nice to disable all caching (ie. Cache on inserts only), and
> > actually get some real numbers as to how the routing algorithm works. What's
> > the point of testing a routing algorithm that can route just about anywhere
> > and hit a cached file?
>
> I have actually simulated this, or at least pretty close to it (I
> simulated only caching when the DataSource is reset, which is 1/30). The
> results were drasticly worse than when caching.
I would not be surprised if overall performance decreases. What I would like
to see is how to get an idea for how the routing algorithm works, and how to
improve it. It is difficult to make any such adjustments when routing is
hidden by caching. I see them as different layers to be optimised
seperately.
> Not caching is most certainly not the right thing to do. We want data to
> be mobile. It is possible that we may need to move away from the "all
Why? Why must data be mobile? I dont follow you. Mobile data is data using
bandwidth unnecessarily. What is your context? What are you actually saying?
I'm not trying to be difficult, I honestly have nooo idea where you are
coming from. :-)
> data is equal" maxim when caching - ie that we prioritize having the
> node keep certain data - but I certain that whether the data arrived by
> DataRequest or InsertRequest is not the right way to weigh that...
*nod* I agree my hypothesis might be wrong, but you have no idea how
overjoyed I am that you seem to be aware of it. I don't currently have the
resources myself, but I hope you have a chance to test it.
Keep well,
Ray
_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://lists.freenetproject.org/mailman/listinfo/devl
