Measurement and Enhancement of Cloud-based Online Gaming Systems

Abstract

Cloud gaming refers to the technologies that offload parts of game software from traditional game consoles to powerful and elastic cloud infrastructure. This design effectively shifts the game system requirements as well as the necessary computational workload to remote cloud platforms and thus has been attracting an increasing amount of attention from both service providers and end users. We have seen commercial cloud gaming systems such as Gaikai and OnLive being available in the market with a considerable user base. However, the pitfalls as well as the design challenges of cloud gaming still remains largely unclear for the general public. In this thesis, we take an initial step towards the understanding of cloud gaming performance in virtualized environments. Different from existing cloud gaming implementations, we migrate the entire gaming system into a fully virtualized local cloud environment and compare its performance to a pure datacenter-based deployment. From our experiments, we find that the virtualized environment will significantly affect the latency as well as the frame rate of the game. As a result, we find the tested games running much slower on the cloud than on their non-virtualized counterpart. In particular, latency increases by about 176 ms and the frame rate decreases by almost 40% in our test platform. To make the matters worse, the resource utilization on the server also increases due to cloud virtualization. To address these issues, we propose a shared-memory-based enhancement to reduce the overhead of games running on virtualized environment. Our evaluation indicates that our approach can successfully reduce the cloud gaming latency by about 6% without noticeable increase of system's CPU and memory utilization.