From:
p2p-hackers@zgp.org (Brad Neuberg)
Date:
Mon Jul 15 20:16:02 2002
Subject:
[p2p-hackers] P2P HTTP
Justin Chapweske wrote:
> Glenn McGrath wrote:
>
>> Everything as a file is a mantra of the unix world. Everything as a file
>> doesnt mean everything is a file, its more like everything has a file
>> handle. For this to be true, the file handle must be independent of the
>> content that it processes.
>>
>> I think the Content Addressable Web (namespace based on content hash)
>> approach used by many systems hinders efforts to see everything as a
>> file
>> becasue the file handle is derived from the hash of the content. It
>> makes
>> every file read-only, if a file is modified it becomes a completely new
>> entry (sometimes this is wanted).
>
>
> The "Content Addressable Web" is a subset, or optimization of generic
> web architecture. It should really only be used in the context of
> generic web architecture, where everything is a URI-addressable web
> object. Indeed, it would be hard to write an application that
> supports only CAW, without supporting the the whole of URIs and web
> architecture, so in practice I don't that anything is hindered in its
> expressiveness.
>
>>
>> If we abstracting the transport layer from our p2p systems in a generic
>> way then individual p2p systems can implement there different
>> search/retrieval system ontop of that, you could have a CAW system ontop
>> of the generic transport layer, maping the hash based filename to an
>> independent file handle.
>>
>> The trasport layer would need properties such as content independence,
>> security, (limited?) anonymity, accountability, intelligent routing...
>> other ?
>>
>
> That generic transport layer already exists, its called HTTP. Add
> SSL/TLS for security, a mix network of HTTP proxies for anonymity, and
> a content delivery network for intelligent routing.
>
> Personally, I think the concept of a P2P transport layer is as vacuous
> as talking about a "P2P operating system".
>
Justin, how would you layer a peer to peer network over HTTP? I know
Magi does something like this, but I'm not completely satisfied with
it's approach. I've been playing around with layering P2P over HTTP
semantics, but it seems like it doesn't really result in any payoff
other than complexity and political correctness. I'm interested in
hearing how you'd do it.
Brad Neuberg