On Thu, 04 Sep 2003, Gordan wrote:

> On Thursday 04 September 2003 08:23, Dan Merillat wrote:
> 
> I understand that. However, unless the border node is heavily CPU bound, that 
> would probably be slower in the long term than having one really big border 
> node and no internal nodes - from the view of the outside, that is. Or am I 
> wrong here? Is proxying a request faster/less intensive than serving? I 
> wouldn't have thought so...

No, to the outside it's not as good as having one huge node.  It is a
chokepoint to all the internal nodes.  From an internal standpoint, you
get faster results since portions of the keyspace end up being
specialized within your LAN/WAN.

Toad: How does the loop detection code work?  Given the hypothetical
situation: Data is outside the lan, in a part of the keyspace that an 
internal node responds quickly to.   If a query hits the gateway
supernode and goes back inside,  is it possible to get back out?

> Yes, but the outside nodes should drop the 10.x routes automatically anyway, 
> as far as I understand.

Yes, but that also drops the keyspace associated with those nodes,
making any data stored inside your lan invisible.  The network will
function, but non-optimally.

> Ideally, you'd block everything outgoing as well, then only enable 
> specifically the ports you want.

Ideally for WHAT?  A prison? Yes.  A corperate server farm? Yes.  A LAN
designated for phone-support personel?  Yes.  An appartment complex
providing high-speed access to the tenants? _NO_NO_NO_.  You can't
always count on a magic firewall to do the blocking for you, since
freenet isn't "well behaved" by internet standards.  You can block SMTP
traffic by blocking port 25, but there's no way to block freenet traffic
short of cutting off all connectivity in both directions.

> But will it really break things that badly? Surely, the nodes will quickly 
> learn to route to the border node(s) instead of trying to route further in. 
> In fact, they won't even bother routing further in because the references 
> will have private IP addresses on them which will get dropped before making 
> it into the routing table in the first place. Or am I wrong here?

Exactly.  But the problem that I'm trying to explain is this:

Outside node 1 searches for key "KA".  Routes it to your gateway,
gateway routes it to internal node 2.  Node 2 responds with a 10.x
address, gets passed through the gateway to 1.  Now 1 has a 10.x noderef
for that keyspace and it gets ignored.  Thus, a successful transfer did
NOT improve routing at all.

If the gateway resets the DS, node 1 gets the data "from" the supernode.
Now the routing table learns that the supernode is a good place to go
for KA.  See the difference?

--Dan

Attachment: pgp00000.pgp
Description: PGP signature

_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to