[ English | Japanese ]
In this research, BitTorrent as one of the most popular and successful P2P applications in thecurrent Internet, is taken as an example in a study of uncoordinated P2P operation.
The first problem to be addressed in this research is how to reveal the topology of real BitTorrentswarms, how dynamic the topology is, and how it affects overall behavior.
We study BitTorrent networks, where real-world BitTorrent swarms were measured using a rigorous and simple method in order to understand the BitTorrent network topology.
We propose usage of the BitTorrent Peer Exchange (PEX) messages to infer the topology ofBitTorrent swarms listed on a BitTorrent tracker claiming to be the largest BitTorrent network on theInternet, instead of building small BitTorrent networks on testbeds such as PlanetLab and OneLabas other researchers have done.
We also performed simulations using the same approach to show the validity of the inferredtopology resulted from the PEX messages by comparing it with the topology of the simulatednetwork.
Our result, verified using the Kolmogorov-Smirnov goodness of fit test and the likelihood ratio testand confirmed via simulation, show that a power-law with exponential cutoff is a more plausiblemodel than a pure power-law distribution.
We also found that the average clustering coefficient is very low, implies the BitTorrent swarms areclose to random networks.
BitTorrent swarms are far more dynamic than has been recognized previously, potentially impacting attempts to optimize the performance of the system as well as the accuracy ofsimulations and analyses.
In the current content delivery architecture, many CDN companies and ISPs adopt hybrid CDN-P2P because of the advantages of P2P.
On the P2P side, peers are organized in a tree-based overlay on a per substream basis for livestreaming.
This ensures that all peers contribute some upload bandwidth.
Each CDN server keeps track of clients currently assigned to it to avoid the undesirable sideeffects of P2P.
Each client learns about other peers assigned to its designed CDN server.
Since in a hybrid CDN-P2P architecture, some of the workload or data delivery is done by peers,CDN servers benefit from a potential reduction in power consumption.
The second problem to be addressed in this research is the trade-off of hybrid CDN-P2Parchitecture compare to CDN.
We solve this problem by proposing a simple model of power consumption of CDN server androuter, including the cost of cooling needed generated from the power consumption of the CDNserver and router.
Furthermore, this power reduction can be used for capacity planning of data centers.
Finally, our proposed methodology can contribute to further characterizing P2P networks andpromotion of relaxing capacity planning data center in term of energy consumption by hybrid CDNP2P.
Keywords: P2P, BitTorrent, Peer-Assisted CDN, Power-Law, CDN, Energy.
Keio University, Graduate School of Media and Governance
MAUI Project
Ph.D. Dissertation
Back to Index Page
ACADEMIC YEAR
2014
NAME
MOHAMAD, Fauzie
TITLE
Analysis of Peer-to-Peer Operation in Content Delivery
ABSTRACT
CONTACT
To obtain the dissertation, please contact;
MOHAMAD, Fauzie ( dikshie at sfc.wide.ad.jp )