Load Balancing in Server Consolidation

Abstract

The growth of server consolidation is due to virtualization technology that enables multiple servers to run on a single platform. However, virtualization may bring the overheads in performance. The prediction of virtualization performance is of especially important. The contribution of our paper is two-fold. First, we propose a general model to predict the performance of consolidation. Second, we study a load balancing problem that arises in server consolidation, where is to assign a number of workloads to a small number of high-performance target servers such that the workloads in each target servers are balancing. We first model the load balancing problem as an integer linear programming. Then, an fully polynomial time approximate scheme (FPTAS) is provided to get the near optimal solution. That is to say, for any given $\varepsilon ≫ 0$, our algorithm achieves ($1+\varepsilon$)-approximation, and its running time is polynomial of both the number of source servers and $1/\varepsilon$ when the number of target servers and the dimensions are constants.