Re: [circle] circle web search wishlist (long)

[Oskar Kosciusko]

In a word, yes. We let one entity own the signposts and everyone is lost.

Nodes should crawl the web, as I don't want the user influencing their own 
crawler or indexer. Otherwise it's all too easy to inject fraudulent data. 
Remember that once people start trusting a system, influencing that system 
becomes profitable, i.e. AdWords. And anyway -- would you really want to 
publish *your* browser cache?

I agree that users should be able to opt in or out of services, and separate 
circles might even make routing easier. If I knew a damn thing about python I'd 
volunteer to help with that. :)

A question -- what exactly determines where a node ends up on the circle? is it 
just a random number that a hacked client could fake, or is it tied to 
something harder to mess with? What would happen to two nodes with exactly the 
same hash key?

oskar

--

--------- Original Message ---------

DATE: Tue, 01 Jul 2003 11:29:25
From: Thomas Voegtlin <address@hidden>
To: address@hidden
Cc: 

>Soooo, you definitely want to remove the Google monopoly?
>That's ambitious...
>
>A few remarks:
>
> - You do not need to have your nodes crawl the web. Simply using the 
>cache of
>your user's web browser should do a better job (and greatly simplify 
>your project)
>
>- Not all circle users will be willing to store 200MB of cache, plus the 
>zillion hashtable
>links that go with it, plus the extra bandwidth use, if they do not use 
>your system.
>
>This points to a more general issue, namely that in the current 
>implementation
>of Circle, all nodes accept to generously store links for everybody. 
>However, some
>people might use chat and not be interested in filesharing, some others 
>just the
>opposite. Why would a chat-only user want to store links to mp3 files? 
>Why would
>the part of the hashtable that contains the list of users on a channel 
>be maintained
>by someone who never uses chat?
>
>(not to mention obvious attacks by saturation)
>
>This generous policy should probably be modified before new services are 
>added
>to the Circle. I can see two possible ways to improve on that:
> 1.- A node should participate only in the services that it uses. This 
>makes the
>most sense, however it would require big changes to the current 
>implementation.
> 2.- A node should maintain a rough balance between the number of links 
>it publishes
>and the number of links it maintains for the other nodes (i.e. 'store 
>link for'). This
>is more compatible with the current implementation, and it would be the 
>first step
>toward detecting attacks by saturation.
>
>
>
>
>
>
>_______________________________________________
>Circle-discuss mailing list
>address@hidden
>http://mail.nongnu.org/mailman/listinfo/circle-discuss
>


____________________________________________________________
Get advanced SPAM filtering on Webmail or POP Mail ... Get Lycos Mail!
http://login.mail.lycos.com/r/referral?aid=27005