Resilient application placement for geo-distributed cloud networks

Abstract

The strong uptake of cloud computing has led to an important increase of mission-critical applications being placed on cloud environments. Those applications often require high levels of availability coupled with guarantees on a minimum level of throughput and a maximum level of response time. To achieve the lowest response time possible, clouds are more and more decentralized, leading to a heterogeneous network of micro clouds positioned on the edge of the network and possibly interconnected by best-effort links. This heterogeneous environment introduces important challenges for the management of these clouds as the heterogeneity results in an increased failure probability. In this paper, we address these challenges by providing a resilient placement of mission-critical applications on geo-distributed clouds. We present an exact solution to the problem, which is complemented by two heuristics: a near-optimal distributed genetic meta-heuristic and a scalable centralized heuristic based on subgraph isomorphism detection. A detailed performance evaluation shows that, with the newly proposed heuristic based on subgraph isomorphism detection, we can double the amount of applications satisfying availability requirements, in cloud environments comprising over 100 nodes, while keeping the time required to calculate the solution under 20s.