young Fedora torrents spread data too slowly

Till Maas opensource at till.name
Sun Oct 12 09:40:33 UTC 2008


On Sun October 12 2008, Mike McGrath wrote:

> Seth a bit about this too (he's our current torrent wrangler) and we
> generally think this was just an issue with two many chunks and not enough
> people for chunks from the beginning.  Which is self correcting but A)
> is annoying and B) is preventable in the future.  We just have to figure
> out the procedure on it.

The question that still needs to be answered is, why torrent1 performed so 
badly. The second seed from Dennis Gilmore only provided 400 KB/s combined 
for every torrent. Then it took around 20 minutes for the majority of peers 
to complete the torrents. So the new seed made it a lot faster than only 
twice as fast. Also it was always visible that the peers exchanged every new 
chunk between them. I do not know which algorithm torrent1 uses to determine 
who it sends some chunks, but it somehow failed. I also noticed that the 
bittorrent protocol in general seems to fail to find the best distribution of 
chunks, because it seems that machines with a high upstream capacity do not 
get faster access to rare chunks or machines with low downstream capacity do 
not connect preferred to machines with low upstream capacity.

It would have been nice to have some details about to how many peers at which 
speeds torrent1 seeded his chunks, because even when it did not choke my 
client, it seeded only with 2 KB/s.
It took around 22 hours for the fastest torrents to distribute around 650 MB 
only from torrent one, which would be a total seeding speed of 8 KB/s.

> Feel free to stop by #fedora-admin with ideas.  It might just be something
> as simple as providing the iso somewhere for a specified time then
> removing, allowing people to seed properly.  I like this because it keeps
> with the community model, allows people to get involved for these sorts of
> releases, and fixes the "slow to start" problem.

There are three issues that should be solved imho:
1) The intial planned seeding should be fast enough
2) It should be easy to workaround any problems regarding 1)
3) Make debugging easier

For 1) I suggest to collect some interested people that are willing to seed 
the Snap1 and to debug the issues with torrent1.

For 2) maybe the isos could be copied were a large number of people around the 
world has access too, e.g. copy it on fedorapeople and give only sponsors 
read access to it.

To improve debugging, there should be statistics on how fast torrent1 seeds 
the new torrents, i.e. how fast to how many peers. This would also allow to 
easily notice if it disbehaves, which seems to have failed yesterday.

Regards,
Till
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 827 bytes
Desc: This is a digitally signed message part.
URL: <http://listman.redhat.com/archives/fedora-devel-list/attachments/20081012/a9bd882f/attachment.sig>


More information about the fedora-devel-list mailing list