ABSTRACT
Almost ten years have passed since the expression "peer-to-peer" reached the large internet
population. It has had considerable success, but there still seems to be a long way ahead.
In these ten years, a wealth of peer-to-peer systems has been proposed by the research
institutions, but only few have made it commercially. While the research community in-
troduces ever more advanced structured peer-to-peer networks, the public seems to prefer
the simplicity and elasticity of unstructured peer-to-peer networks, despite their numerous
and obvious flaws. Even developers or system designers seem to prefer either unstructured
networks, or, at best, hybrid systems. Why?
One of the important factors, business considerations aside, is that designing, imple-
menting and testing one such system before deployment is virtually impossible for some-
one without a PhD in the area. Even comparing existing systems is difficult, because there
is no unifying benchmarks or models.
This thesis addresses this problem by introducing theoretical models for both struc-
tured and unstructured networks, by implementing simulators to help system designers get
a better idea of the performance before going to a real implementation and, finally, by
showcasing an image retrieval application on structured peer-to-peer networks.
We organize this work in 4 chapters, in addition to the introductory and concluding
remarks. First, we make a thorough overview of the state of the art in peer-to-peer networks,
both structured and unstructured, as well as existing attempts at providing a generic model
for such networks. In the same chapter, we also review techniques in information retrieval
on peer-to-peer networks, as well as image retrieval techniques that we will subsequently
adapt for our application environment. Afterwards, we dedicate one chapter for each of the
three important contributions of the thesis: the structured network model, the unstructured
network model and the image retrieval application.
This thesis demonstrates the possibility of defining new peer-to-peer networks or mod-
eling existing designs easily, as well as the applicability of image retrieval techniques in apeer-to-peer environment with scarce resources.