Mike McGrath wrote:

On Sat, 11 Oct 2008, Denis Leroy wrote:


Mike McGrath wrote:

I'm not saying something isn't wrong. I'm saying I'm not seeing it and if
I can't see it I don't know how I'm going to fix it. Can anyone give me
speeds for any other torrent besides ours?

Well I just added the Fedora 9 i686 live CD for comparison, I'm getting 1.0
MBytes per second.



So, on a new, less popular torrent like the Snap1 you see slow speeds. On
an older, more popular torrent you see faster speeds. How is the torrent
not behaving as it should again?


As someone else mentioned, the distribution of the chunks pushed to the
clients is supposed to be randomized. Apparently it isn't, we're all
sitting here with the same completion percentage and the exact same
chunks, all trying to download from the same one seed.


I'm not sure whether it's the client that picks the random chunk to
download, or the server that imposes it though. Maybe we're all using
the same buggy client :-) (I'm using transmission).


Can you add an extra seed using a different client ? (you can use
transmissioncli for example).


--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

On Sat, 11 Oct 2008, Ralph Angenendt wrote:

> Mike McGrath wrote:
> > I just downloaded:
> >
> > http://torrent.fedoraproject.org/torrents//F10-Snap1-x86_64-Live.torrent
> >
> > It took just over 6 minutes and my top speed was 2037.3 KB/s I might
> > suggest that different places have different rules wrt torrent and qos,
> > firewalls and other setups can affect torrent preformance.
>
> I wonder how you did that, as I see exactly 1 peer which has 100% (the
> rest is at around 88%), and that is torrent1.fedoraproject.org. This doesn't
> seem to be a performance issue >
>

I'm testing from another host in the EU right now and I'm getting
somewhere around 3090.3 KB/s. I'm all for quicker torrent downloads so if
y'all want to stop by #fedora-admin and chat it up perhaps we can come up
with a fix or better guidelines on how to use torrent or something.

-Mike

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

On Sat, 11 Oct 2008, Denis Leroy wrote:

> Mike McGrath wrote:
> > On Sat, 11 Oct 2008, Denis Leroy wrote:
> >
> > > Mike McGrath wrote:
> > > > I'm not saying something isn't wrong. I'm saying I'm not seeing it and
> > > > if
> > > > I can't see it I don't know how I'm going to fix it. Can anyone give me
> > > > speeds for any other torrent besides ours?
> > > Well I just added the Fedora 9 i686 live CD for comparison, I'm getting
> > > 1.0
> > > MBytes per second.
> > >
> >
> > So, on a new, less popular torrent like the Snap1 you see slow speeds. On
> > an older, more popular torrent you see faster speeds. How is the torrent
> > not behaving as it should again?
>
> As someone else mentioned, the distribution of the chunks pushed to the
> clients is supposed to be randomized. Apparently it isn't, we're all sitting
> here with the same completion percentage and the exact same chunks, all trying
> to download from the same one seed.
>

Explain the timeline on how that happens to everyone, at the same time,
all of us getting every other chunk except this one, and none of us has
completed it (except me and my two machines)?

-Mike

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

Mike McGrath wrote:

On Sat, 11 Oct 2008, Denis Leroy wrote:


Mike McGrath wrote:

I'm not saying something isn't wrong. I'm saying I'm not seeing it and if
I can't see it I don't know how I'm going to fix it. Can anyone give me
speeds for any other torrent besides ours?

Well I just added the Fedora 9 i686 live CD for comparison, I'm getting 1.0
MBytes per second.



So, on a new, less popular torrent like the Snap1 you see slow speeds. On
an older, more popular torrent you see faster speeds. How is the torrent
not behaving as it should again?


My download roared along at 700 to 800 KB/s until I had the same ~71% that
everyone else had, then my download speed dropped to 1 to 8 KB/s. The
leechers are happily sharing everything they've got, but everyone has the
same chunks. The one seed is passing out "virgin" chunks at about 5 KB/s.
At that rate, it'll take another 24 hours before anyone has the whole
thing, at which point _everyone_ will have it.

When I look at the peer list, the only people getting more than 8 KB/s
are those with less than 71% completion. They can receive chunks from
many sources. Everyone else has to wait for the one seed to send out
something new.

I'm running bittorrent-4.4.0.

--
Bob Nichols "NOSPAM" is really part of my email address.
Do NOT delete it.

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

Mike McGrath wrote:

On Sat, 11 Oct 2008, Denis Leroy wrote:


Mike McGrath wrote:

On Sat, 11 Oct 2008, Denis Leroy wrote:


Mike McGrath wrote:

I'm not saying something isn't wrong. I'm saying I'm not seeing it and
if
I can't see it I don't know how I'm going to fix it. Can anyone give me
speeds for any other torrent besides ours?

Well I just added the Fedora 9 i686 live CD for comparison, I'm getting
1.0
MBytes per second.


So, on a new, less popular torrent like the Snap1 you see slow speeds. On
an older, more popular torrent you see faster speeds. How is the torrent
not behaving as it should again?

As someone else mentioned, the distribution of the chunks pushed to the
clients is supposed to be randomized. Apparently it isn't, we're all sitting
here with the same completion percentage and the exact same chunks, all trying
to download from the same one seed.



Explain the timeline on how that happens to everyone, at the same time,
all of us getting every other chunk except this one, and none of us has
completed it (except me and my two machines)?


Here's my theory:

At the border of someone's network there is a gateway that has been set
up to restrict the bandwidth of torrent packets severely. There is a
community of users that can access the seed without going through that
gateway. They can access the seed and share chunks at high speed. The
rest of the world, however, can only see the dribble of data that is
allowed through that gateway.

The only solution is to use another protocol to load a site on the
other side of that (unknown) gateway with a full seed.

--
Bob Nichols "NOSPAM" is really part of my email address.
Do NOT delete it.

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

On Sat October 11 2008, Mike McGrath wrote:

> I'm not saying something isn't wrong. I'm saying I'm not seeing it and if
> I can't see it I don't know how I'm going to fix it. Can anyone give me
> speeds for any other torrent besides ours?

You can fix it by adding a seed to the group of peers that do not get chunks
fast from any other seed.

Regards,
Till
--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

On Sat October 11 2008, Mike McGrath wrote:

> I'm testing from another host in the EU right now and I'm getting
> somewhere around 3090.3 KB/s. I'm all for quicker torrent downloads so if
> y'all want to stop by #fedora-admin and chat it up perhaps we can come up
> with a fix or better guidelines on how to use torrent or something.

We need more seeds at the beginning and more people with access to the isos.
Dennis Gilmore finally managed to acquire the isos somehow and seeded them
and this made it possible for everybody else to download them, too.

I guess you did not finish a download of the Snap1 isos, because the tracker
did not show while, but they were shown for Dennis and everyone else who
finished then. Therefore the testing procedure for the torrents also needs to
be improved. You can give the files to some people who are willing to seed
them some hours before they are published and see, whether they can really
get them and whether the tracker also counts the downloads. Then you also
have more initial seeders.

Regards,
Till


--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

John Reiser wrote:

Young torrents on http://torrent.fedoraproject.org/ spread data too slowly.
At 0500 UTC on Sat.Oct.11, there was 1 seed and 14 leeches, all at 13%
done.
At 1300 UTC on Sat.Oct.11, there was 1 seed and 31 leeches, all at 55%
done.

This is for F10-Snap1-x86_64-Live.
I turned my machine off for the night (US PDT), powered on in the morning,
and within half an hour had caught up to all leeches at 57% done.
I use tranmsissionbt, and I saw that the seed was sending pieces of 256 KiB
to me at 2.5 KiB/s, taking almost two minutes per piece (and then
"pausing.")


About 2230 UTC on Sat.Oct.11, the 67 leeches finished within a few minutes
of each other. The torrents started about 0000 UTC. So it took more than 22 hours
to distribute the 67 copies of 726MB of data. That is approx. 10KB/s for each,
on average, which is very slow. The expected speed is at least 10 times as fast.

One theory is deliberate throttling/choking somewhere.
Here's a [current] traceroute for two of my machines that were in the swarm
during those 22 hours [first few hops omitted due to paranoia]:
-----
traceroute to torrent.fedoraproject.org (152.46.7.220), 30 hops max, 38 byte packets
[server in internet hotel]
3 ge-6-0-0.edge-rtr02.ptldor02.iinet.com (198.145.40.99) 0.427 ms 0.424 ms 0.394 ms
4 fe-6-0-0-20.acs-rtr04.ptldor02.iinet.com (198.145.240.165) 0.855 ms 0.652 ms 0.521 ms
5 66-193-100-89.static.twtelecom.net (66.193.100.89) 1.711 ms 2.106 ms 1.613 ms
6 peer-01-so-0-0-0-0.snjs.twtelecom.net (64.129.248.17) 16.834 ms 49.743 ms 17.190 ms
7 te3-2.mpd01.sjc01.atlas.cogentco.com (154.54.6.81) 223.951 ms te8-2.mpd01.sjc01.atlas.cogentco.com (154.54.6.237) 74.034 ms te8-2.ccr02.sjc01.atlas.cogentco.com (154.54.6.233) 17.179 ms
8 te8-4.mpd01.lax01.atlas.cogentco.com (154.54.2.129) 63.993 ms te8-4.ccr02.lax01.atlas.cogentco.com (154.54.2.149) 38.244 ms 35.766 ms
9 te2-8.mpd01.iah01.atlas.cogentco.com (154.54.5.101) 75.458 ms te9-8.mpd01.iah01.atlas.cogentco.com (154.54.0.230) 264.798 ms te9-8.ccr02.iah01.atlas.cogentco.com (154.54.3.185) 63.395 ms
10 te4-2.ccr01.atl01.atlas.cogentco.com (154.54.7.237) 77.549 ms te9-2.ccr01.atl01.atlas.cogentco.com (154.54.24.1) 106.957 ms 78.105 ms
11 te8-2.mpd01.atl04.atlas.cogentco.com (154.54.3.174) 89.485 ms te7-1.mpd01.atl04.atlas.cogentco.com (154.54.6.114) 77.320 ms te7-3.mpd01.atl04.atlas.cogentco.com (154.54.6.122) 92.799 ms
12 MCNC.demarc.cogentco.com (38.99.223.154) 86.542 ms 86.692 ms 86.470 ms
13 torrent1.fedoraproject.org (152.46.7.220) 86.432 ms !<10> 119.191 ms !<10> 86.265 ms !<10>
-----
traceroute to torrent.fedoraproject.org (152.46.7.220), 30 hops max, 60 byte packets
[home ADSL]
4 fa0-13.na02.b006468-1.pdx01.atlas.cogentco.com (38.103.165.25) 49.514 ms 51.751 ms 51.966 ms
5 gi4-1.3506.core01.pdx01.atlas.cogentco.com (38.20.33.201) 53.016 ms 53.243 ms 53.924 ms
6 po10-0.core01.sfo01.atlas.cogentco.com (154.54.3.133) 68.408 ms 58.018 ms 58.898 ms
7 te3-1.mpd01.sfo01.atlas.cogentco.com (154.54.3.102) 61.053 ms 57.317 ms 57.872 ms
8 te7-4.mpd01.sjc01.atlas.cogentco.com (154.54.6.134) 60.721 ms 60.586 ms 58.419 ms
9 te8-4.mpd01.lax01.atlas.cogentco.com (154.54.2.129) 70.422 ms 69.758 ms 70.212 ms
10 te2-8.mpd01.iah01.atlas.cogentco.com (154.54.5.101) 106.394 ms te3-8.mpd01.iah01.atlas.cogentco.com (154.54.2.197) 105.619 ms 104.902 ms
11 te9-2.ccr01.atl01.atlas.cogentco.com (154.54.24.1) 118.926 ms 117.071 ms 118.478 ms
12 te7-1.mpd01.atl04.atlas.cogentco.com (154.54.6.114) 117.427 ms te7-3.mpd01.atl04.atlas.cogentco.com (154.54.6.122) 118.588 ms te8-4.mpd01.atl04.atlas.cogentco.com (154.54.3.170) 119.276 ms
13 MCNC.demarc.cogentco.com (38.99.223.154) 130.398 ms 126.942 ms 127.524 ms
14 torrent1.fedoraproject.org (152.46.7.220) 128.325 ms !X 128.997 ms !X 129.545 ms !X
-----

--

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

On Sat, Oct 11, 2008 at 06:16:27PM -0700, John Reiser wrote:
> About 2230 UTC on Sat.Oct.11, the 67 leeches finished within a few minutes
> of each other. The torrents started about 0000 UTC. So it took more than 22 hours
> to distribute the 67 copies of 726MB of data. That is approx. 10KB/s for each,
> on average, which is very slow. The expected speed is at least 10 times as fast.
>
> One theory is deliberate throttling/choking somewhere.
> Here's a [current] traceroute for two of my machines that were in the swarm
> during those 22 hours [first few hops omitted due to paranoia]:

I'm pretty sure this was throttling done by torrent1.fedoraproject.org
itself. I was coming in via Internet2, and I assume the reverse route
was also via Internet2 (which might be a bad assumption).

4 WORCESTER-BOSTON.GODDARD.GIGAPOP.NET (130.215.6.2) 1.947 ms 2.015 ms 2.041 ms
5 nox1sumgw1-Vl-591-NoX-WPI.nox.org (192.5.89.41) 2.111 ms 2.092 ms 2.470 ms
6 nox300gw1-Vl-803-NoX.nox.org (192.5.89.238) 2.991 ms 2.642 ms 2.679 ms
7 nox300gw1-PEER-NoX-INTERNET2-192-5-89-222.nox.org (192.5.89.222) 7.883 ms 7.856 ms 7.712 ms
8 ge-2-3-1.0.rtr.atla.net.internet2.edu (64.57.28.19) 12.137 ms 11.702 ms 11.662 ms
9 so-0-0-0.0.rtr.atla.net.internet2.edu (64.57.28.6) 25.262 ms 25.396 ms 25.401 ms
10 dep7600-gw2-internet2.ncren.net (198.86.17.65) 33.395 ms 33.404 ms 33.439 ms
11 rtp7600-gw-to-dep7600-gw2.ncren.net (128.109.70.137) 35.270 ms 35.507 ms 36.060 ms
12 rtp11-gw-to-rtp7600-gw-sec.ncren.net (128.109.70.121) 36.187 ms 36.056 ms 36.068 ms
13 ibiblio-gw-to-rtp11-gw.ncren.net (128.109.70.114) 36.363 ms 36.371 ms 36.509 ms
14 torrent1.fedoraproject.org (152.46.7.220) 35.207 ms !X 35.207 ms !X 35.268 ms !X

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list

On Sat, Oct 11, 2008 at 09:24:28PM -0400, Chuck Anderson wrote:
> On Sat, Oct 11, 2008 at 06:16:27PM -0700, John Reiser wrote:
> > About 2230 UTC on Sat.Oct.11, the 67 leeches finished within a few minutes
> > of each other. The torrents started about 0000 UTC. So it took more than 22 hours
> > to distribute the 67 copies of 726MB of data. That is approx. 10KB/s for each,
> > on average, which is very slow. The expected speed is at least 10 times as fast.
> >
> > One theory is deliberate throttling/choking somewhere.
> > Here's a [current] traceroute for two of my machines that were in the swarm
> > during those 22 hours [first few hops omitted due to paranoia]:
>
> I'm pretty sure this was throttling done by torrent1.fedoraproject.org
> itself. I was coming in via Internet2, and I assume the reverse route
> was also via Internet2 (which might be a bad assumption).

Looks symmetrical to me.

traceroute to 130.215.6.2 (130.215.6.2), 30 hops max, 40 byte packets
1 152.46.7.254 (152.46.7.254) 0.915 ms 1.139 ms 1.359 ms
2 rtp11-gw-to-ibiblio-gw.ncren.net (128.109.70.113) 0.248 ms 0.238 ms 0.215 ms
3 rtp7600-gw-to-rtp11-gw-sec.ncren.net (128.109.70.122) 0.364 ms 0.382 ms 0.427 ms
4 dep7600-gw2-to-rtp7600-gw.ncren.net (128.109.70.138) 3.401 ms 3.452 ms 3.534 ms
5 internet2-to-dep7600-gw2.ncren.net (198.86.17.66) 19.154 ms 19.150 ms 19.195 ms
6 ge-0-1-0.10.nycmng.abilene.ucaid.edu (64.57.28.7) 23.927 ms 23.909 ms 23.908 ms
7 ge-1-3-0.666.rtr.atla.net.internet2.edu (64.57.28.18) 28.981 ms 28.940 ms 29.070 ms
8 nox300gw1-Vl-110-NoX-INTERNET2.nox.org (192.5.89.221) 33.709 ms 42.725 ms 48.957 ms
9 nox1sumgw1-Vl-803-NoX.nox.org (192.5.89.237) 33.796 ms 33.829 ms 33.856 ms
10 nox1sumgw1-PEER-NoX-WPI-192-5-89-42.nox.org (192.5.89.42) 33.988 ms !X 34.042 ms !X 34.080 ms !X


--
Matt Domsch
Linux Technology Strategist, Dell Office of the CTO
linux.dell.com & www.dell.com/linux

--
fedora-devel-list mailing list
fedora-devel-list@redhat.com
https://www.redhat.com/mailman/listinfo/fedora-devel-list