A publication of the

Abstract:

Current cloud computing infrastructure typically assumes a homogeneous collection of commodity hardware [1]. Some applications may have a high computational processing requirements that cannot be obtained from these commodity servers. These include numerical intensive applications, applications with high parallelism, and applications that exhibit near real-time performance requirements. One strategy for increasing processing efficiency to the cloud is to add heterogeneity; providing access to resources better suited to complex computation [2]. Heterogeneous systems integrate more than one processing unit with different performance and energy consumption characteristics. Unlike its homogeneous counterpart, a heterogeneous system address both throughput and efficiency for various workloads by matching resources to each application’s needs and has a much higher potential in saving energy. However, exploiting this potential requires a well designed resource allocation system that maps heterogeneous resources to applications’ requirement with minimum cost in performance and power. This is a nontrivial task. Field-programmable gate array (FPGAs) offer a significant acceleration in execution over CPUs. They are also significantly more power-efficient, resulting in a computational efficiency improvement in the orders of magnitude over both CPUs and GPUs [3],[4]. As more and more workloads are being deployed in the cloud, it is appropriate to consider how to make FPGAs and their capabilities

Año de publicación:

Keywords:

Fuente:

google