Hi there,

First off, great work on 1.2dev! Despite the "experimental" nature of it, it runs quite well!

I've noticed a bit of a problem, though. I'm in a situation where there are over 20 leechers, and one very slow seed that only seems to have a maximum of one outbound connection. Basically, one leecher gets some parts of the file, and then "seeds" them out to the rest of the swarm.

What I have been noticing is that, in this slow swarm, KTorrent is always dropping most of the peers and only staying connected to a leech or two. Since the leeches are usually overloaded or on slow connections, I tend to lag behind the swarm quite a bit.

If I stop and then resume the torrent, I am almost immediately caught up to the rest. KTorrent finds a fast leech and quickly downloads the difference.

I'm wondering why KTorrent is dropping most of the peers. It seems to me that it would be much faster if it keeps looking at all of them. Is this some method to reduce overhead from BitTorrent protocol messages or something? Pausing and resuming is not too bad, but I'm wondering if the peer-dropping is an intended behaviour.

I'm running a SVN version checked out a few days ago. So, fairly recent!

We only actively drop connections when we :
- shutdown KT or stop or delete the torrent
- ownload is finished, we disconnect from all seeders
- ban a peer

So this is kind of weird.

I'm not sure but aren't we dropping connections from peers that are choked for a long time?

We did at one point in time but that code is no longer used, and that was only when there was an upload limit.

I have no limits set.

It does seem like peers are being dropped when they've been choked for a while...