Optimizing performance of batch of applications on cloud servers exploiting multiple GPUs

The cloud computing is an economical paradigm to access high processing power at low price. In this context, cloud servers are strongly loaded in order to reduce user costs. GPUs offer very interesting performances that can boost many applications. Mixed with cloud computing, it can propose a way to cope with the data deluge. The server side of the cloud is a great playground for GPUs as they can run at full capacity, and enable a better energetic efficiency and a lower environment impact. In order to propose many user services, cloud servers host many virtualized operating systems. Our work is focused on proposing an integrated a management system to share multi-GPUs resources between many virtualized user applications. This paper will focus on performance improvement when many applications are simultaneously using GPUs.