Most of BitTorrent systems contains implementation of an algorithm predicting the download time of the torrent. Commonly used Simple ETA algorithm is known for its vulnerability on download rate fluctuations and lack of precision. Many authors address the topic of torrent evolution prediction, but majority of created models focus on torrent's global parameters or takes unrealistic assumptions and therefore are unusable in ETA forecast. In this thesis we analyze the conditions influencing the torrent download and propose the local-view model for ETA prediction. We analyze the model qualities and validate it against the real-life torrent traces. We show that created model improves ETA prediction quality when compared to Simple ETA algorithm.
@MastersThesis{,
author = {Piotr Powa{\l}owski},
title = {Local View Modeling for {ETA} in {BitTorrent}},
school = {Vrije Universiteit},
address = {Amsterdam, The Netherlands},
year = {2008},
month = aug
}