Resource allocation algorithm for GPUs in a private cloud

摘要

Cloud computing is currently playing an important role in different applications. Deploying graphical processing units (GPUs) as a resource in clouds constitutes a major challenge. As the number and the scale of internet services increase, the services running on a cloud computing environment also increase. These services could have been efficiently improved in terms of a service level agreement (SLA) if we could make use of the GPUs architectures in the cloud infrastructure. Cloud providers currently offer GPUs as an infrastructure as a service (IaaS). Providing GPUs in another context such as software as a service (SaaS) requires a resource allocation algorithm. This paper proposes a resource allocation algorithm (RAA) for GPUs in the cloud environment. The proposed RAA will provide a way for cloud providers to offer GPUs in a SaaS context. The RAA is based on the load balancing mechanism trying to fully utilise all the GPUs resources. This paper also investigates the suitability of the RAA by conducting a video enhancement service on a private cloud. It shows that the proposed algorithm provides scalability and enhancements in terms of a SLA.