bt.ban_ratio setting

[MaubyMan]

In order to make uTorrent more aggressive in banning bad peers should I increase or decrease this value? I have searched these forums and read differing suggestions on this matter.

The manual defines this setting as: The lowest acceptable ratio of good to bad pieces a peer can send before it gets banned.

Here is how I interpret this with the following settings:

bt.ban_ratio = 128 (default)

bt.ban_threshold = 1 (my setting)

bt.ban_use_ban_ratio = true

If a peer sends 100 good pieces and 2 bad ones, the good to bad ratio for that peer is 100/2 = 50.

Since bt.ban_threshold is surpassed (2 bad pieces), and good/bad ratio is less than bt.ban_ratio (50<128), the peer would be banned.

So lowering bt.ban_ratio to 49 or less would make the peer safe.

Therefore it is my opinion that decreasing bt.ban_ratio makes uTorrent more tolerant and increasing it makes uTorrent more aggressive.

Am I correct?

[Ultima]

That's correct.

[MaubyMan]

Thanks for verifying.

[Firon]

That being said, µTorrent keeps track of which peer sent which chunks. When it successfully downloads a piece, it verifies which peer sent the bad chunks on that same piece and gives a high equivalent of hashfails. I don't remember the exact amount, but it basically makes ban ratio less useful than expected.

[MaubyMan]

µTorrent keeps track of which peer sent which chunks

By 'chunks' do you mean what uTorrent calls 'blocks'? (the 16KB sections that make up a piece)

and gives a high equivalent of hashfails

Not sure what you mean there. Did you mean for example if a peer sends 2 bad chunks in one piece, then it is credited with 2 hashfails?

On another note, wouldn't it be more efficient to just re-download the bad chunks/blocks instead of the entire piece? For example, on a torrent with 4MB pieces (256 16KB blocks), if a peer sends 1 bad chunk/block, thus rendering the piece invalid, it is obviously easier to just re-download 16KB instead of 4MB. Is there something else you have to take into account?

[Firon]

There are no hashes in the torrent for the 16KB chunk, only for the full piece. That's why the entire piece has to be thrown away and re-downloaded.

And what I mean is that when it marks a peer for a verified hashfail (by comparing the calculated hashes of the blocks from the bad piece and the good piece), it counts as if it were more than 1 hashfail. I'm not sure of the exact number.