On Mon, 28 Feb 2005 12:03:07 +0000, Tony Garnock-Jones <tonyg at lshift.net>  
wrote:

> Okay. I'll have to do a bit of digging to find out how messages are  
> routed to the two hits - 30 seconds seems like a loooooong time, so I  
> guess I need to learn more about how the p2p network works to understand  
> why 30 seconds isn't practically infinity.
>
It is. If the thing works, I found it much faster than SqueakSource in  
responsiveness and comparable to an average remote repository.

> In fact I'll, just as an experiment, try setting it to 300 seconds and  
> see what happens.
>
Nothing. I think the nodes that claimed to have the file you wanted simply  
weren't available. That's why network size matters here. And stable code,  
of course ;)

> Agreed :) How does bittorrent do its thing, I wonder? I did read the  
> protocol description once many moons ago but it didn't stick, clearly...
>
Bittorrent asks lots of peers for blocks, I think. With direct  
connections, there's no peer-to-peer routing of payload as far as I'm  
aware. Gnutella does p2p routing of "metadata" (searches and responses),  
but not of data - actual files are transported out-of-band with direct  
HTTP-ish connections. Something I'd like to postpone as long as possible  
;) (I think for large files, say >>100k, things should be splitup in  
blocks and one could retrieve individual blocks).

Anyway, I think that the mod where you ask for a 'file handle' or  
'transfer handle' to a peer first is probably the most sensible thing to  
change on the current implementation.