Raphael Manfredi wrote:
> Quoting Christian Biere <[EMAIL PROTECTED]> from ml.softs.gtk-gnutella.devel:
> :I'm not sure what's meant with "exploit". GWebCaches have always been
> :free for all and for a very long time, Gtk-Gnutella did not have its
> :own bootstrap servers. It's actually a huge leap backwards that we
> :seem to need client-specific bootstrap servers. It shows that a
> :minority of egoistic and incompetent minds is successfully destroying
> :the spirit anf infrastructure of an once open and vendor-independent
> :protocol and network.

> I just want to ensure a viable fallback for GTKG peers should the GWC
> system fail one day.

It's certainly a good idea to have one or more GTKG-only caches as
a last resort but I don't really think you need Fort Knox for that.
You can use a usual GWC with some minor incompatibility to normal
GWebCaches and make sure that GTKG doesn't submit its URL to other
GWebCaches. Since Gtk-Gnutella is not that wide-spread, abuse is
easily detectable. The port hopping is easy to implement. I don't
think you need any obfuscation here and it should be sufficient
that some bozo puts the URL into some bootstrap list. And even if
he would, there's nothing running on the hardcoded port 90% of the
time which saves a lot of traffic.
 
> The UHC system uses far less bandwidth than GWCs today, but it is
> unusable by UDP-firewalled clients, and it is also probably less (ab)used
> than GWCs currently.

> But perhaps the solution is not to build a GTKG-specific server.  Perhaps
> the solution is to create a binary protocol over TCP that would use
> much less b/w than the current GWC system?

I don't really think the traffic per request is that important nor
does the exact protocol matter. It's the policy that matters. DNS
is often praised for being robust but if everybody banged the root
servers as Gnutella peers do with GWebCaches, it would certainly
collapse as well. Of course in this case ISPs could simply configure
their routers to treat port 53 specially (which they probably do
anyways) which is something that won't happen for Gnutella or
GWebCaches.

Anyway, I would suggest to get rid of updates to caches. Caches
should only send (URL) updates to each other. For clients this
should be taboo. IP address updates are unnecessary due to UHCs
as well and TCP crawling is possible too - although it's far
less efficient due to the properties of TCP. However, this let's
the cache control more of its bandwidth and does actually
scale.

> With only ~20 GWCs worldwide and the amount of traffic generated, the
> current system is doomed in the long run.

I'm not so sure about this. Servers and traffic keep getting cheaper
and cheaper. These few servers can and do handle much more than
100 something that existed a few years ago. In German there's a
saying "Totgesagte leben laenger" which seems to fit the GWebCache
system quite well. It's been called "dead" almost since the beginning.
Of course 20 sitting ducks are a serious issue.

-- 
Christian

Attachment: pgpZZGpxZlkC9.pgp
Description: PGP signature

Reply via email to