Year of Publication


Degree Name

Doctor of Philosophy (PhD)

Document Type

Doctoral Dissertation




Computer Science

First Advisor

Dr. Zongming Fei


In this dissertation we take a peer-to-peer approach to deal with two specific issues, fair trading and file distribution, arisen from data management for cloud computing.

In mobile cloud computing environment cloud providers may collaborate with each other and essentially organize some dedicated resources as a peer to peer sharing system. One well-known problem in such peer to peer systems with exchange of resources is free riding. Providing incentives for peers to contribute to the system is an important issue in peer to peer systems. We design a reputation-based fair trading mechanism that favors peers with higher reputation. Based on the definition of the reputation used in the system, we derive a fair trading policy. We evaluate the performance of reputation-based trading mechanisms and highlight the scenarios in which they can make a difference.

Distribution of data to the resources within a cloud or to different collaborating clouds efficiently is another issue in cloud computing. The delivery efficiency is dependent on the characteristics of the network links available among these network nodes and the mechanism that takes advantage of them. Our study is based on the Global Environment for Network Innovations (GENI), a testbed for researchers to build a virtual laboratory at scale to explore future Internets.

Our study consists of two parts. First, we characterize the links in the GENI network. Even though GENI has been used in many research and education projects, there is no systematic study about what we can expect from the GENI testbeds from a performance perspective. The goal is to characterize the links of the GENI networks and provide guidance for GENI experiments.

Second, we propose a peer to peer approach to file distribution for cloud computing. We develop a mechanism that uses multiple delivery trees as the distribution structure, which takes into consideration the measured performance information in the GENI network. Files are divided into chunks to improve parallelism among different delivery trees. With a strict scheduling mechanism for each chunk, we can reduce the overall time for getting the file to all relevant nodes. We evaluate the proposed mechanism and show that our mechanism can significantly reduce the overall delivery time.