On Virtualization of Reconfigurable Hardware in Distributed Systems

In the design of next-generation distributed and high-performance computing systems, Reconfigurable Processing Elements (RPEs) such as FPGAs and multi-core heterogeneous computers will play an important role. FPGAs are well-known for their programmability, power efficiency, reasonably high performance, functional flexibility, and their scalability. However, proper virtualization schemes at higher software abstraction layers to utilize these RPEs are under research. In this paper, we propose a virtualization framework for distributed computing systems that supports RPEs. First, we present various scenarios in terms of use-cases to discuss the utilization of RPEs in distributed computing systems. Secondly, we propose a scheme to virtualize RPEs in distributed systems. Based on various virtualization levels, we provide a general model for a computing node which incorporates both General Purpose Processors (GPPs) and RPEs. Thirdly, we present a typical application task model. Finally, we present a case study of a large-scale application from the bioinformatics domain, which demands different types of processing elements in a distributed computing system.