Evaluation of GPU Virtualisation Approaches for Machine Learning Enhanced Debugging of Cloud Orchestration

Abstract

Nowadays, computing demand on General-Purpose Graphics Processing Units (GPGPUs) is steadily increasing due to the great interest in machine learning. The computational time of embarrassingly parallel tasks can be reduced with such GPUs by orders of magnitude compared to CPUs. In this paper, we briefly overview a wide range of GPU virtualisation strategies (including API remoting, para/full virtualisation and hardware based virtualisation) and their related methods. The fundamental details are also discussed to understand the differences between the presented solutions. Finally, the key features are described and are evaluated to help choose an effective baseline framework for a challenging graph-based machine learning method to be applied in the field of debugging of cloud orchestration.