On the benefits of the remote GPU virtualizationmechanism: The Rcuda case

摘要

Graphics processing units (GPUs) are being adopted in many computing facilities given theirrnextraordinary computing power, which makes it possible to accelerate many general purposernapplications from different domains. However, GPUs also present several side effects, such asrnincreased acquisition costs as well as larger space requirements. They also require more powerfulrnenergy supplies. Furthermore,GPUs still consume some amount of energy while idle, and theirrnutilization is usually low for most workloads. In a similarway to virtualmachines, the use of virtualrnGPUs may address the aforementioned concerns. In this regard, the remote GPU virtualizationrnmechanism allows an application being executed in a node of the cluster to transparently use thernGPUs installed at other nodes. Moreover, this technique allows to share the GPUs present in therncomputing facility among the applications being executed in the cluster. In this way, several applicationsrnbeing executed in different (or the same) cluster nodes can share 1 ormore GPUs locatedrnin other nodes of the cluster. Sharing GPUs should increase overall GPU utilization, thus reducingrnthe negative impact of the side effects mentioned before. Reducing the total amount of GPUsrninstalled in the cluster may also be possible. In this paper, we explore some of the benefits thatrnremote GPU virtualization brings to clusters. For instance, this mechanism allows an applicationrnto use all the GPUs present in the computing facility. Another benefit of this technique is thatrncluster throughput, measured as jobs completed per time unit, is noticeably increased when thisrntechnique is used. In this regard, cluster throughput can be doubled for some workloads. Furthermore,rnin addition to increase overallGPUutilization, total energy consumption can be reduced uprnto 40%. This may be key in the context of exascale computing facilities, which present an importantrnenergy constraint. Other benefits are related to the cloud computing domain, where a GPUrncan be easily shared among several virtual machines. Finally,GPUmigration (and therefore serverrnconsolidation) is one more benefit of this novel technique.