[bittorrent] Peer and blocks asking behaviour
dwknight at depthstrike.com
Wed Jul 30 18:43:56 EDT 2008
Given the difficulty in causing a collision in SHA1 with a known piece
size, I don't think collisions are THAT big of an issue immediately.
The single-peer filling actually makes it harder for one peer to cause a
multitude of hashfails. If you get a single piece entirely from a single
peer and it fails, you can ban the peer on ONE hashfail without risking
cross-contaminating the sample data.
Thad Ward wrote:
> On Wed, Jul 30, 2008 at 07:45:36PM +0200, Adrian Ulrich wrote:
>> My client tries to fill pieces with data from a single peer:
>> This might be slower than asking multiple clients but makes it harder
>> for an 'evil' peer to corrupt multipile pieces.
> I think getting all the chunks for a single piece from the same peer
> makes it easier for an "evil" peer to corrupt the data, as it could
> (in theory at least) send a piece that is a hash collision with the
> correct data. Getting the data from multiple peers makes it harder for
> one peer to corrupt the swarm, provided there are more non-evil peers
> than evil peers that have the correct data.
> BitTorrent mailing list
> BitTorrent at lists.ibiblio.org
More information about the BitTorrent