A cluster-based decentralized job dispatching for the large-scale cloud.

Abstract

The remarkable development of cloud computing in the past few years, and its proven ability to handle web hosting workloads, is prompting researchers to investigate whether clouds are suitable to run large-scale computations. Cloud load balancing is one of the solution to provide reliable and scalable cloud services. Especially, load balancing for the multimedia streaming requires dynamic and real-time load balancing strategies. With this context, this paper aims to propose an Inter Cloud Manager (ICM) job dispatching algorithm for the large-scale cloud environment. ICM mainly performs two tasks: clustering (neighboring) and decision-making. For clustering, ICM uses Hello packets that observe and collect data from its neighbor nodes, and decision-making is based on both the measured execution time and network delay in forwarding the jobs and receiving the result of the execution. We then run experiments on a large-scale laboratory test-bed to evaluate the performance of ICM, and compare it with well-known decentralized algorithms such as Ant Colony, Workload and Client Aware Policy (WCAP), and the Honey-Bee Foraging Algorithm (HFA). Measurements focus in particular on the observed total average response time including network delay in congested environments. The experimental results show that for most cases, ICM is better at avoiding system saturation under the heavy load.