high performance computing on graphics processing units: hgpu.org

In this paper we propose MR-API framework, a comprehensive framework Application Programming Interface for Heterogeneous Multi-core systems in a cloud environment. Recent trends have shown the significant importance of adopting Heterogeneity in Multi-cored processing environment due to their potential high performance and energy efficiency. In order to extract the maximum potential of underlying hardware, the API must be efficient enough such that it efficiently automates the process of mapping the computation to underlying computational resources. However, most of the approaches proposed rely on the programmer to specify the mapping manually, such an approach demands a lot of work to be done by a programmer. Hence MR-API, a novel API framework that adopts efficient heuristics in automating the process of mapping computations to processing elements which also includes map-reduce in order to exploit the explicit parallelism of the application in a cloud environment. By adopting MR-API framework to map computations to processing elements on a CPU+GPU nodes in cloud environment, it is also established that the proposed API framework is robust and adaptable to various benchmark sets and system configurations.