Neil Barsema wrote on 8/1/00 10:25 am:

>Hello,
>
>I've read a couple of times 
>that Freenet has a problem 
>with node discovery . 

I put up a proposal for dealing with this not too long ago.  It got ignored in 
a mess of "Napster going down" messages.  I've decided to wait until after 0.3 
is released before reposting.  Check throug the archives over the last week for 
message.

>However this interpreted 
>much to broadly, All you 
>need to use Freenet is one 
>initial node and the 
>discovery after that is 
>actually handled pretty 
>elegantly using the actual 
>datasource as a reference 
>not the node the request 
>was ent to.
>(I remeber it well when Ian 
>pointed out this feature). So 
>you'll accumalate  list of 
>nodes as you use the 
>network. No problem there.

The only problem is that if that one starting node is malicious, you quite 
possibly never will get to a non-malicous connection without  tarting over.  My 
prnposal took care of this by allowing the user to quickly build up a long list 
of nodes from many diffrent sources.  Combined with your plan, I think my 
proposal will turn out to be the best option.

>
>The problem Freenet has 
>with node discovery is that 
>'new' nodes don't get 
>incorporated into the 
>network, as in no requests 
>will be routed towards them. 
>This means all the juicy 
>resources the new nodes 
>offer aren't used. That is 
>what the Inform.php is for, 
>to 'anounce' new nodes. Not 
>to discover nodes in the 
>network.

Broadcasts were considered, but they're way too open to attack.  I think your 
system is much better.

I have done some actual codeing for my proposal, and at this point it is ready, 
except for debug+testing (I haven't even compiled it yet because Kaffe is 
totaly fscked on my box).  I don't think your idea would be too difficult to 
add.  Its just a matter of having the server remember the IP of the client when 
it connects.  The other details of your plan should be carried out elsewhere.

>
>I think this could be 
>addressed by tweaking the 
>way Freenet chooses the 
>refernce in the datastore a 
>bit.
>
>When servicing a request the 
>node knows of 4 locations 
>the data can be found.
>- node 1 the datasource
>- node 2 the node the reply 
>is comming from  node 3 the 
>node itself
>- node 4 the node it 
>forwarded the reply to.
>
>We already implemeted a 
>random possibillity that the 
>first and the seccond node 
>are the same this done by 
>node 2 by resetting the 
>datasource to itself.  suggest 
>we remove this and move 
>this functionallity from 
>where it resets the 
>datasource to where the 
>reference is set 
>(DataSend.java?). The 
>possibillities could be 
>something like 90% node 1  
>8% node 2 and 2% node
>4; (I guess we need to 
>simulate)
>
>This would mean that there 
>is a 2% chance your node will 
>be 'discovered' if you make a 
>request so all you need to do 
>to get incorperated into 
>freenet is make enough 
>requests.
>
>
>                                               
>                                       Neil
>
>
>
>
>
>_____________________________
>__________________ 
>Freenet-dev mailing list
>Freenet-dev at lists.sourcefor
>ge.net
>http://lists.sourceforge.net/
>mailman/listinfo/freenet-de
>v 


-----------
Go Ping Yourself

_______________________________________________
Freenet-dev mailing list
Freenet-dev at lists.sourceforge.net
http://lists.sourceforge.net/mailman/listinfo/freenet-dev

Reply via email to