Combining Peer-to-Peer and Cloud Computing for Large Scale On-line Games

This thesis investigates the combination of Peer-to-Peer (P2P) and Cloud Computing to support Massively Multiplayer On- line Games (MMOGs). MMOGs are large-scale distributed applications where a large number of users concurrently share a real-time virtual environment. Commercial MMOG infras- tructures are sized to support peak loads, incurring in high economical cost. Cloud Computing represents an attractive solution, as it lifts MMOG operators from the burden of buy- ing and maintaining hardware, while offering the illusion of infinite machines. However, it requires balancing the tradeoff between resource provisioning and operational costs. P2P- based solutions present several advantages, including the in- herent scalability, self-repairing, and natural load distribu- tion capabilities. They require additional mechanisms to suit the requirements of a MMOG, such as backup solutions to cope with peer unreliability and heterogeneity. We propose mechanisms that integrate P2P and Cloud Computing combining their advantages. Our techniques allow operators to select the ideal tradeoff between performance and economical costs. Using realistic workloads, we show that hybrid infrastructures can reduce the economical effort of the operator, while offering a level of service comparable with centralized architectures.