Uni-Logo
Sie sind hier: Startseite Lehre Aktuelle Lehre Oberseminar CoNe (Computer Networks) Statistical Behavior of Peer-to-Peer Networks
Artikelaktionen

Statistical Behavior of Peer-to-Peer Networks

Mrs. Shyamala Villupuram Sundararaman (Diploma Thesis)

Abstract

The paper discusses node’s behavior for predicting the next status of node in the BitTorrent nerwork using existing predictors namely SatCounter-2 predictor which uses 2-bit saturat- ing counter for prediction, history predictor, hybrid predictor, rightnow predictor, twiddled history predictor and linear predictor. Existing predictors assume that the peers are always online and predicts status of node based on last online/offline state of node using saturating counter values. New predictor predicts node’s availability in the network by finding prob- abilities of measurement of node’s status based on last online/offline state of node instead of saturating counter. When the probability is high it is possible to predict the node will be online, otherwise offline. Predict node’s availability status for next 12 hrs and showing the result in de bruijn graph. The new predictor uses three techniques for predicting node’s availability where these three techniques are implemented using Markov Predictor. The paper explains existing predictors and implementation of existing predictors and also the three techniques of new predictor in detail and also produce corresponding results for both the predictor. Then, the paper discuss advantages of these new techniques and also how to implement the new predictor. The paper further compares results of existing predic- tors with the new predictor, the time and space complexity of the existing predictors and new predictor and future works on new predictor is also presented in order to improve the new predictor’s performance. Finally, the outcome acquired from the new predictor will be provided as result.

Benutzerspezifische Werkzeuge