Profit-based file replication in data intensive cloud data centers

摘要

Many of the applications running in cloud data center are data intensive, processing large amount of data inside the data center. File replication, which brings data files closer to the computing virtual machines (VMs), is an effective strategy that reduces data access latencies and bandwidth consumption, thus saving energy in data centers. In this paper, we formulate and study the file replication problem (FRP) in data center, with the goal of minimizing the total energy consumption of data file access inside data centers. In contrast to all the existing work of data replication in data centers, which are mainly heuristic based, we design a time-efficient approximation algorithm with performance guarantee for energy consumption in file replication. In particular, our file replication algorithm is based on a novel concept called “profit”, and optimizes over a submodular function that can be computed efficiently. Our algorithm yields the total profit of file replication at least half of what is achieved by an optimal replication solution. We also design two energy- and time-efficient heuristic file replication algorithms. Via extensive simulations using CloudSim, a popular simulation framework for cloud computing, we compare all the algorithms under different network scenarios. We show that the approximation algorithm outperforms the other two under different network parameters, while all three effectively reducing the total energy consumptions of data access in data centers.