> We are using the current version of yum that comes with Fedora Core 2 > (2.0.7). However, our changes to the actual yum code are very minor > (less than 30 or so lines I would say) so all changes should be forward > and backward compatible. We also included one extra file to handle the > BitTorrent connections. There are some extra files that implement the > background uploading and an intelligent server manager that helps to get > the swarms started. > Our implementation basically just hooks into urlgrabber. The > headers.info file is modified on the server to point to a .torrent file > instead of a .rpm file. When url grabber detects that it is downloading > a .torrent, it trips our code. Our code downloads the .torrent normally > then starts the BitTorrent process which continues with the download. > When the download is complete, yum contacts our client upload manager > and tells it that there is another file to seed. Yum then continues > normally and, with the exception of some extra debug output, the user > and yum can't tell and don't care if the file was downloaded via HTTP or > BT. Yum completes its work normally, but the client manager runs in the > background continuing to upload on the swarms that can use it. So you setup a separate seed for each file? Do you keep the seed around after the download completes? Aren't you taking a fairly serious performance hit? -sv
