| We define a framework to implement multi-party applications in which cooperation between application peers is leveraged to achieve scalability. In this framework, the peers cooperate by devoting a part of their own resources to be used by other peers of the same application group. Through detailed analysis, simulations, and implementations, we demonstrate that applications implemented using this approach can achieve significant performance benefits.; There are two major aspects to scalably implement cooperative applications---resource provisioning and management, and handling the impact of peer failures on application performance. In this thesis we define efficient techniques to address both these aspects.; Resource limitation is the most typical reason that constrain the scalability of applications. Therefore we first study resource provisioning and management at application peers. We do this using two examples. The first example application is the group key distribution problem that arises in secure group communication systems. Our proposed solution to this problem is based on the cooperative approach and is, therefore, fundamentally different from all existing solutions to this problem. We perform detailed comparisons with best-known existing techniques to demonstrate that our proposed scheme incurs the least processing, storage and communication overheads. As a second example we examine the one-many data transfer problem. We proposed a scalable solution to this problem using a cooperative approach known as application-layer multicast. This solution constructs provably efficient data delivery structures, guarantees low control overheads, and is the most scalable application-layer multicast protocol known today.; Next we study the impact of peer failures on application performance. While the cooperative model for multi-party applications enables scalability, it adds dependencies between application peers. Consequently failures of some peers affect the performance at other peers. Efficient failure recovery techniques, therefore, form an important component of cooperation-based implementations of multi-party applications. We examine this aspect of cooperative applications using the application-layer multicast example. We define a simple, robust, low-overhead data recovery technique that efficiently handles failures of application peers. (Abstract shortened by UMI.)... |
