Lioux launched a BitTorrent campaign.
He persuaded me to install the bittorrent port on jim, and donate 6KB/s (ca. 40% of my upstream bandwith) to his project. Additional i have set a limit of 6 connections, calculating that each connection should get 1KB.
After running one night, on next morning my nagios was crying because “the Internet is slow”. I checked with Hendriks excellent slurm program and noticed that my upstream was blowing out full speed (16KB/s, yes my ADSL line is slow).
I checked with tcpdump what caused this traffic and immediately noticed several wellknown hosts as primary offenders:
10:49:17.908038 IP 21322530218.direct.eti.at.53481 > g164193.upc-g.chello.nl.acmsoda: F 1347279866:1347279866(0) ack 1734755715 win 33304
... 10:50:05.497482 IP 21322530218.direct.eti.at.52573 > panda.droso.net.6881: P 1:49(48) ack 1 win 33304 10:50:05.545402 IP 200-140-3-148.bsace703.dsl.brasiltelecom.net.br.54871 > 21322530218.direct.eti.at.7241: R 3804027892:3804027892(0) win 0 10:50:05.606741 IP 21322530218.direct.eti.at.7241 > creme-brulee.marcuscom.com.51748: FP 0:7(7) ack 1 win 33304 0:50:05.607486 IP 21322530218.direct.eti.at.54575 > foo.lovett.com.7241: . ack 1 win 33304
. Obviously lioux persuaded some other fellow commiters (kwm, marcus, erwin and ade) to run BT as well and now we exchange our distfiles via BT (hooray!).
It looks like the traffic shaping algorithm is far from optimal and according to netstat the maximum connection number was also violated. And why is so much traffic required to coordinate the peers, as we all have the relevant files and don’t download anything?
My conclusion, currently bittorrent can’t compete with fetch(1), ftp(1) and an established mirror hierachy when it comes to distributing free software.