[BitTorrent] Some complement about Tracker-tracker communication
justin at street-vision.com
Mon Feb 21 10:23:37 EST 2005
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> On windows, the number of connections you need to use to reach the
> limit is around the number of connections required for 2000-5000
> peers with a best-seen of a little under 10,000 peers (I know it's
> not many, but it seems that the windows TCP/IP stack doesn't like
> having dozens-thousands of half-closed connections lying around).
> On linux, the number of connections you need to use to reach the
> limit is around the number of connections required for 1,000,000+
> peers, although it may be less because of other limitations such as
> routers (if I remember right, we have yet to see a single tracker
> actually reach these sort of limits, most trackers are hitting
> interest limits in the 300k-400k range).
> UDP _MAY_ solv some of the problems involved here, but there doesn't
> seem to be enough acceptence by the development community for it to
> be useful as a solution to this problem.
Well it sounds like less of an issue given what you say, if the answer
is just not to use Windows for your tracker if you want to scale.
How many half closed connections is that roughly? Could probably make
an estimate from the bandwidth I suppose.
UDP ought to scale a bit better, although if there is a lot of packet
loss this could end up not being the case. Especially as tcp send is
very efficient, and they are small messages unlikely to be fragmented
anyway, may not make much difference under a good implementation.
> When I hit the 10,000 peer limit of Windows, the bandwidth used
> (pre-compact implementation on my tracker) was less than 10kbyte/sec.
Thats very little bandwidth
So compact really doesnt save much, no point in trying to introduce
a compact for for ipv6 as some have suggested.
> - -----Original Message-----
> From: Justin Cormack [mailto:justin at street-vision.com]
> Sent: Monday, February 21, 2005 5:34 AM
> To: BitTorrent at yahoogroups.com
> Subject: Re: [BitTorrent] Some complement about Tracker-tracker
> > Tracker-Tracker communication already has some implementation in
> > real-world environments.
> > BNBT based trackers running on versions 7.3, 7.7, and 8.0 have the
> > capability (with the consent of the administrators of both
> > trackers) to share ALL peer data.
> > I personally operate a 5-way (not including the dedicated hub)
> > linked tracker network. There are indeed several steps to take at
> > the start to insure proper peer data sharing, but most have been
> > documented at http://cbtt.depthstrike.com/trackerlinks.
> > Additionally, deployment guidelines have been documented there as
> > well.
> Just a few questions about these:
> I see the main point is not listed as being bandwidth (despite what
> people say about the huge importance of compact=1 suggesting that
> are bandwidth limited) or even reliability (the solution has a single
> of failure tracker hub apparently too, rather than peer to peer
> solutions, though it will add some extra reliability), but number of
> This is a bit surprising to me as the limiting factor, how many
> are we talking about? And doesnt UDP tracker solve this by being
> connectionless? Or are there other problems with udp, or is it just
> widely used?
> > I have been working on a php implementation of automated editing of
> > tracker information within .torrent files and returning them to
> > clients (in an effort to simplify my own tracker network's use).
> > - -----Original Message-----
> > From: guanying_wang [mailto:guanying_wang at yahoo.com.cn]
> > Sent: Saturday, February 19, 2005 7:21 AM
> > To: BitTorrent at yahoogroups.com
> > Subject: [BitTorrent] Some complement about Tracker-tracker
> > communication
> > The torrent file contains tracker address and file hash, which are
> > the
> > only useful infomation torrent file contains. Trackers can identify
> > different torrent file by checking file hash, and Merkle Tree hash
> > is short enough for quick identification. If we can turn to wide
> > deployment of Merkle Tree as soon as possible, I think
> > tracker-tracker
> > communication will be very easy to implement.
> It doesnt make much difference whether you use Merkle tree or info
> they are the same size.
> > Scalability is good, because capacity can effectively be improved
> > by simply adding additional trackers.
> > Thank you.
> > Guanying
> > PS: I haven't seen my first post till now. God knows when I can,
> > maybe
> > tomorrow... How do you guys use this list? I think the web
> > interface is not good enough. RSS, either. Thank you.
> There are moderation delays sometimes, on all interfaces. I use the
> version, havent seen the first post either...
> - --
> No virus found in this outgoing message.
> Checked by AVG Anti-Virus.
> Version: 7.0.300 / Virus Database: 266.2.0 - Release Date: 2/21/2005
> -----BEGIN PGP SIGNATURE-----
> Version: PGP 8.0.3
> -----END PGP SIGNATURE-----
> Yahoo! Groups Links
Yahoo! Groups Links
<*> To visit your group on the web, go to:
<*> To unsubscribe from this group, send an email to:
BitTorrent-unsubscribe at yahoogroups.com
<*> Your use of Yahoo! Groups is subject to:
More information about the BitTorrent