[vox-tech] bittorrent - no seeds but distributed copies increase

[Peter Jay Salzman]

On Mon 26 Jul 04,  3:45 PM, Rob Rogers <@.net> said:
> On Mon, Jul 26, 2004 at 02:46:48PM -0700, Peter Jay Salzman wrote:
> > On Mon 26 Jul 04,  2:37 PM, Samuel N. Merritt <spam at andcheese.org> said:
> > > On Mon, Jul 26, 2004 at 02:08:22PM -0700, Peter Jay Salzman wrote:
> > > > Question:
> > > > 
> > > > How does the "distributed copies" get larger when there are no seeds?
> > > 
> > > I think that "distributed copies" measures how many complete copies of
> > > the file you could get if you took all the pieces that everyone has and
> > > assembled them. 
> > > 
> > > For example, consider a five-part file.
> > > Alice has: 1 2 3 4
> > > Bob has:       3 4 5 
> > > Carol has: 1 2   4 
> > > 
> > > You could make one complete copy of the file from all this, so there'd
> > > be 1 distributed copy. If Carol got piece 5 from Bob, then you could
> > > assemble two complete copies. 
> > > 
> > > That's the integer part of distributed copies; I'm not sure where the
> > > fractional part comes from. Maybe it's the size of the largest
> > > distributed incomplete subfile divided by size of file, but that's just
> > > a shot in the dark. 
> 
> IIRC, that's exactly how it works. So in the above example you would
> have 1.8 distributed copies. (Alice's 4 peices and Bob's #5 is one.
> Carol's 1 and 2, and Bob's 3 and 4 make another, and we ignore the last
> Carol's last peice because we alread counted Bob's)
> 
> > ok.  this was my understanding.
> > 
> > > > Does the tracker ever inject packets into the torrent when needed (like
> > > > when seeds == 0 and distributed copies < 1.0)?
> > > 
> > > No. The tracker doesn't have a local copy of the file. If there are no
> > > seeds and < 1 distributed copy, everyone's download will stall before
> > > finishing 
> > 
> > this was also my understanding.  but my question still stands: how does
> > the distributed copies increase if there are no seeds?
> > 
> > i'm looking at a bittorrent right now.  it's remained constant at:
> > 
> >    seeds: 0 seen now, plus 0.983 distributed copies
> >    peers: 19 seen now, 98.4% done at 0.2 kB/s
> > 
> > that ".983 distributed copies" has been creeping upwards.  last i looked
> > at it, about 15 minutes ago, it was at .97.   i've noticed this happen
> > before too.
> > 
> > how exactly does that number increase when there are no seeds?
> > 
> > pete
> 
> In some newer BT clients there's an option for a mode called superseed.
> If you activate superseed mode (usually only used by the original
> seeder) you won't appear as a seed to the tracker, even though you have
> a full copy of the torrent. It also activates a more intelligent seeding
> mode. Clients will know which peices are available from other clients,
> and will usually try to request the most rare peice the find to try to
> increase the number of distributed copies faster. In superseed mode, a
> seed will tell the tracker it has no peices. When a client connects to
> the superseed, it will tell the client it has just received a peice (and
> I believe tell the client that is the only peice it has) forcing the
> client to only request that peice. Once that peice has been downloaded
> by the client, the superseed will not offer another peice to that client
> until it sees another client with that peice (meaning that the first
> client had sent it on)
 
nice.  i actually wondered why i never heard of seeders giving
preferences to rare packets.  it seemed like an obvious thing to do.

thanks for explaining that -- not only does it make perfect sense, it
also answers what (used to be) a mysterious question i've had for
months.   :-)

thanks rob!

pete

-- 
In theory, theory and practise are the same.  In practise, they aren't.
GPG Instructions: http://www.dirac.org/linux/gpg
GPG Fingerprint: B9F1 6CF3 47C4 7CD8 D33E 70A9 A3B9 1945 67EA 951D