Calantha: Content Distribution Across Geo-Distributed Datacenters

Large cloud service providers often replicate data to multiple geographically distributed datacenters for availability and service quality purposes. The enormous amount of data needed to be shuffled among datacenters call for efficient schemes to maximally exploit the capacity of the inter-datacenter networks. In this paper, we propose Calantha, a new rate allocation scheme that improves the reliability and operability of content distribution across geo-distributed datacenters, without sacrificing capacity utilization and max-min fairness among competing sessions. Calantha leverages hop-constrained spanning tree to enhance the reliability of inter-datacenter links. A novel approximation algorithm is proposed to solve the rate allocation problem in polynomial time and achieve alpha-optimal approximation. Our simulation results have shown that we can reduce the number of spanning trees by 4 4.5 %, as well as has 2.7 % more average capacity utilization and 1.0 % less minimum spanning tree calculations.