Measurement Based Research On The Characteristics Of K-bucket Against The EMule/Kad Network

Peer-to-peer network is not only an emerging as well as promising technology today but also the hot topic in the academic world. DHT, namely distributed hash table, plays the role as the cutting-edge tech in peer-to-peer structured network. eMule is one of those that adopt DHT technology and widely deployed across the globe, with its Kademlia (one algorithm of DHT) implementation, Kad. However, recent research on Kad network mainly focus on the macro-characteristics, while few of them show any interest in Kad's routing table, k-bucket, on which this thesis is concentrating. Based on the latest and updated research achievement in this field, the eMule client software is modified for our measurement purpose. At the mean while, a framework for statistics and analysis fully packed with scripts which were written for automation. The framework itself is highly configurable, customizable and extensible, providing a solid foundation for future's research work.Then the characteristics of both space and time dimension is represented. The time dimension characteristics of k-bucket consist of session length, inter-arrival, downtime and lifespan. According to the CCDFs, contacts in the k-bucket behave quite different from the so-called well known, heavy-weight or power-law distributions, but fit into the exponential distribution category very well. After splitting into separate arrival Poisson flows, hyper-exponential distribution can well depict and model the characteristics accordingly.The space dimension characteristics of k-bucket include physical geographic distribution and ID hash space distribution. The former is found to be identical to that of the whole network, revealing that there is no preference against the geographic location. The user behavior in China is such different from other European countries in the means that they only appear in Kad in a short period of time although the concurrent user population is the largest, which is bad for the stability and development of the Kad network.In the aspect of ID distance, k-bucket is found to be having a strong "distance effect". For instance, the distant k-bucket is always full with contacts and their average session length is always high, whereas those adjacent ones not. "Distance effect" with eMule/Kad, revealing the inconsistency between routing entries, is a new finding which can be contributed to improve the routing algorithm and the performance of network. With further experiments and investigations, the mechanism could be well simulated based on the model of M/M/k/k queuing system. As the simulation result shows, the selection mechanism does not affect the residual life in k-bucket. However, the undetermined period, which is from the moment that peer goes off, to the time that its unavailability's detected, might possibly be the root cause of "distance effect".