On demand service provisioning on Peer-to-Peer networks

Abstract

Abstract The demand for distributed applications has been increasing since the birth of internet. It has scaled geographical areas in search of information and computational resources for processing. The recent developments in areas like the grid, cloud and utility computing have enabled researchers in need of compute power to utilize resources from a globally shared pool. The emergence of Service Oriented Architectures and Web Services has contributed to the development of several platforms for grid/cloud computing which offer a new way to create loosely-coupled dynamic distributed systems. Thus introduction of Virtual Organizations has added an impetus towards distributed applications by providing on-demand service provisioning. The use of job-based paradigms and strong coupling of current service-based paradigms with static registries such as UDDI hinder the achievement of complete dynamism over volatile resources of the distributed frameworks. A possible solution is the use of structured peer to peer overlay networks, which has emerged as a means of sharing data and computing power where the nodes act as peers of each other to keep a check on the resources as well as handle the volatility of the system. In this thesis, the architecture of a demand-driven web service deployment framework is presented which allows sharing of data and computing capacity using p2p technology as its backbone. Thus the use of dynamic peer-to-peer (p2p) techniques within a web service based framework introduce the ability of the network to adapt to resource volatility which has been already established in p2p-based content-delivery models. Consideration of a service as well as user speci cation of resources for provisioning consumer requests increases performance of the entire architecture. The proposed framework also incorporates a proper load balancing approach between the servers, thereby increasing the utilization of resources in the networks. One of the main focus of this architecture is decentralization of the registry. The distributed registry in the proposed architecture has been implemented in such a way that it makes service discoverable form any part of the network, increases the service availability, and provides a better platform for handling the scalability of the framework. The use p2p le sharing techniques within this framework reduces the overhead of fetching the deployable code from the service repository and sharing it among the deployed instances to carry out successive deployments. Thus, this thesis focuses on various issues such as resource availability, scalability and abstraction. Demand-driven resource allocation is based on request parameters and availability of the resources to create the basis for a fully dynamic virtual market place of computational resources.