BitTorrent Traffic Characteristics

Abstract

This paper reports on a measurement and modeling study of session and message characteristics of BitTorrent traffic. BitTorrent is a Peer-to-Peer (P2P) replication and distribution system developed as an alternative to the classical client-server model to reduce the load on content servers and networks. Results are reported on measurement, modeling and analysis of application and link layer traces collected at the Blekinge Institute of Technology (BTH) and a local ISP in Sweden. Link layer traces and application logs were collected, modeled and analyzed using a dedicated measurement infrastructure developed at BTH to collect P2P traffic. New results are reported on important session and message characteristics of BitTorrent, i.e., session interarrivals, sizes and durations, request rates and response times. Our results show that session interarrivals can be accurately modeled by a second-order hyper-exponential distribution while session durations and sizes can be reasonably well modeled by various mixtures of the Log-normal and Weibull distributions. Response times have been observed to be modeled by a dual Log-normal mixture, while request rates are modeled as dual Gaussian distributions.

Lyapunov stability and performance of user-assisted Video-on-Demand services

“…Several mathematical models and experiments confirm its resiliency and robustness [12][13][14][15][16][17][18][19][20][21]. The models developed by Laurent Massouli and Milan Vojnović are similar in spirit than the ones of Yang De Veciana [22], and Qiu-Srikant [15], via coupon replication systems, with rigorous mathematical results [23].…”

On the Double-Faced Nature of P2P Traffic

“…As their results are comparable with the ones reported in this paper, we consider promising our approach, which doesn't rely on the protocol opcode information to distinguish download from signaling traffic. In another related work [2], the authors have analyzed BitTorrent traffic characteristics, using an instrumented version of the official client. They suggest that conversation durations and sizes can be reasonably well approximated by the lognormal distribution, which is also confirmed in our results.…”

Scalable onion routing with torsk

“…Since currently there are around one million (= 2 20 ) concurrent Kad nodes and roughly 2 11 ORs, it would be difficult to find an OR simply by searching for random nodeIDs. For this reason, in our implementation Torsk nodeIDs have a specific format: nodeID = r1||torskID1||r2||torskID2, where r1 and r2 are random strings, and torskID1 and torskID2 are strings common to all Tor ORs.…”

“…The various message sizes were measured empirically. The simulations also include churn of both clients and routers; the churn model for routers was derived empirically from a 1000-hour survey of the Tor network, while the churn model for clients was estimated using arrival rates measured by [19] and session times measured by [11]. The resulting networks had a client to router ratio of approximately 60 : 1.…”