Embedded Processor Virtualization for Broadband Grid Computing

We implemented and evaluated a heterogeneous system architecture that combines a traditional computer cluster with a broadband network of embedded set-top box (STB) devices to provide a distributed computing platform for parallel applications. Our prototype system for broadband grid computing leverages the recent dramatic progress in computational power of STBs. It includes a complete head-end cable system based on the Tru2way standard, a DOCSIS-2.0 network, and an implementation of the Open MPI library running on the STB embedded operating system across 128 devices. An important contribution of our work is a novel method for the virtualization of a large collection of embedded processors within a managed broadband network. This enables the embedded processors to transparently inter-operate with servers in the computer cluster using the message-passing model. To evaluate the interoperability, performance, and scalability of our system we completed a set of experiments with the standard IMB MPI benchmark suite as well as two real parallel applications. The experimental results confirm that there is an important convergence trend between traditional computing and embedded computing and that a broadband network of embedded processors is a promising new platform for a variety of computationally-intensive and data-intensive grid applications.