--- Toad <[EMAIL PROTECTED]> wrote: 
> On Fri, Oct 24, 2003 at 11:34:50AM +0200, Some Guy wrote:
> >  --- Toad <[EMAIL PROTECTED]> wrote: 
> > > On Thu, Oct 23, 2003 at 02:12:55PM +0200, Some Guy wrote:
> > > > Was NGR tested as extensively as freenet's original routing?  If not then why?
> > > > 
> > > > You could have reused some of the same code from the first freenet paper.  
> > > > Even now this 
> > > might
> > > > still be a good idea.  Sorry if you guys have done this and I haven't heard, 
> > > > but I never
> was
> > > all
> > > > that convinced by NGR.
> > > 
> > > Nope. Ian was dazzled by its beauty. But the main problem was that it's
> > > virtually impossible to simulate NGRouting.
> > 
> > Hmmmm, dazzled by its beauty eh.  
> > 
> > I don't see why testing it would be any harder than testing the old routing as was 
> > done in the
> > freenet paper(freenet.pdf 1999).  For the paper the simulated 1000 nodes.  They 
> > didn't have
> move
> > actual data around.  The NG 10 points seems like less state per node to simulate 
> > than the old
> > system.
> 
> It's not 10 points. It's 16 IIRC. But that's per estimator and we have a
> bunch of estimators and a bunch of decaying averages per node, which are
> then combined to get an estimate for a given key, which we use to route.
> But the big issue is that it's virtually impossible to simulate it because
> it involves timings and is intentionally highly sensitive to network
> conditions.

Ok, you're right its a couple points per vertex.  That's still not a lot of state.  
Most of the
simulation's memory will probably be storing keys in caches (not the data though).

I'm not talking about a serious simulation of:
* network latency
* congestion
* packet loss
* faulty nodes
* bandwidth limitations
* nodes disconnecting
* nodes requesting things by taste
* random bits of alchemy
* reconnection costs
* node resource overload
The first freenet paper ignored these too.  I'm just suggesting a simular simulaton 
that shows if
nodes use NG to estimate hop count and pick their optimal route, the system will 
specialize and be
efficient.  

If NG works for a homogenious random graph of nodes that never disconnect with 
unlimited
bandwidth, we can all feel a bit more confortable about it.  If this isn't case, 
well... we might
need to figuar out why.


> > Boy it sounds like I'm volunteering aren't I!  I'm not as dazzled by its beauty 
> > though.  Still
> if
> > nobody has time to do this, maybe I should.  Is there any interest in writing a 
> > paper about NG
> as
> > for the original routing?  Does anyone else think this is a good idea???
> 
> There was one. You have read it? On the project web site.
You're talking about this right?:
http://www.freenetproject.org/index.php?page=ngrouting
It's nice.  It gives you an idea of how NGR should work.  It might be nice to have a 
paper that
shows that it works.

Maybe I can figuar out what O(f(N)) NG routes in, or what O(g(N)) it takes for a 
compeletly
nonspecialized net to converge.  Can you think of anything else that should be looked 
for?

I can think of one interesting experiment where you take some functions and see how 
well NG fits
to them, maybe the routing time of an idea Plaxton Routing node -- an upside down 
spike. 

__________________________________________________________________

Gesendet von Yahoo! Mail - http://mail.yahoo.de
Logos und Klingelt�ne f�rs Handy bei http://sms.yahoo.de
_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to