Optimal dataset allocation in distributed heterogeneous clouds

Abstract

We consider optimal datasets allocation in distributed Cloud Computing systems. Our objective is to minimize processing time and cost. Processing time includes virtual machine processing time, communication time, and data transfer time. In distributed Cloud systems, communication time and data transfer time are important component of processing time because data centers are distributed geographically. If we place datasets far from each other, this increases the communication and data transfer time. The cost objective includes virtual machine cost, communication cost, and data transfer cost. Cloud service providers charge for virtual machine usage according to usage time of virtual machine. Communication cost and transfer cost are charged based on transmission speed of data and dataset size. The problem of allocating datasets to VMs in distributed heterogeneous Clouds is formulated as a linear programming model with two objectives: the cost and processing time. After finding optimal solutions of each objective function, we use a heuristic approach to find the Pareto front of multi objective linear programming problem. In the simulation experiment, we consider a heterogeneous Cloud infrastructure with five different types of Cloud service provider resource information, and we optimize dataset placement by guaranteeing Pareto optimality of the solutions.