[GNUnet-developers] Re: gnunet 0.4.0

[Christian Grothoff]

On Tuesday 11 June 2002 02:21 am, you wrote:
> On Tue, 11 Jun 2002, Christian Grothoff wrote:
> > I've just checked how much traffic is really generated. It looks like
> > that everything is ok - except that you may want to reduce the number of
> > hosts that you're connected to. The number is by default 128, and since
> > there are not that many hosts online yet, GNUnet tries to connect *a
> > lot*, trying to saturate its connections. So most of the traffic that you
> > see is actually just an attempt to establish connections. (at least on my
> > machine).
>
> I left it running for the night, got 2 gigs of udp traffic. ;) I'll see
> about changing the number of connections next (and try to modify
> trafshow to separately report incoming and outgoing counters). If that
> doesn't help, I'll try take a closer look at the debug output and
> your code.

I found something. In 'isPending', we were forwarding a query if we 
a) had not seen it before or
b) had seen and forwarded it before but the TTL expired before the
     new TTL that we got this time.

This allows a query to be forwarded by us multiple times if the network is 
large enough (as opposed to my testing networks); the reason is, that in the 
first round:

a=>b
a=>c

and then 

c=>b

which would usually be detected (ttl smaller) but due to the *randomness* in 
GNUnet, this does not always detect that it is the same query. The fix is 
simple - drop clause 'b'. If we have forwarded the query (and it is still 
live in our routing table), don't forward. 

I've put the fix into the CVS version (the code is a bit more complicated, 
but this the idea). So if you don't like 2 GB of traffic (*very* large 
anonymity set :-), better get the CVS version :)

Christian