Date Added: Jul 2011
This paper introduces a peer-to-peer framework for providing, locating and consuming distributed services that are encapsulated within virtual machines. The authors believe that the decentralized nature of peer-to-peer networks acting in tandem with techniques such as live virtual machine migration and replication facilitate scalable and on-demand provision of services. Furthermore, the use of virtual machines eases the deployment of a wide range of legacy systems that may subsequently be exposed through the framework. To illustrate the feasibility of running distributed services within virtual machines, several Hadoop benchmarks are executed on a compute cluster running the framework, and their performance characteristics are evaluated. While I/O-intensive benchmarks suffer a penalty due to virtualization-related limitations in the prevailing I/O architecture, the performance of processor-bound benchmarks is virtually unaffected.