Placement of applications in computing clouds using Voronoi diagrams

Abstract

Abstract The vision of millions of users launching tens of millions of applications running on millions of globally scattered servers presents a new challenge for Cloud Providers: how to assign so many virtual applications to physical servers, while meeting latency needs, improving network utilization, and satisfying availability constraints. Today’s application placement puts too much burden on the cloud user, lacks scalability and inhibits the global reach of Public Clouds. The size, breadth, and dynamic nature of Public Clouds present a special challenge to the task of placement. Cloud Providers able to provide rapid decisions and frequent optimizations for placement will have significant competitive advantage. Not only will they provide the best customer experience, their costs will be lower as they make more efficient use of their network resources. However, given the calculations required, data structures and the algorithms used to process location-based decisions must be as globally scalable as the Public Clouds themselves. In our study, we define a novel data structure, the Virtual Cloud Model, for modeling global cloud resources. We adapt a well-known geometric device, Voronoi Diagrams, and combine it with near-real-time network latency information. We then solve the application placement problem and suggest an API for Cloud Providers to support both “low-latency” and “high-availability” applications. The algorithms are scalable, parallelizable and distributable.