MoKa: A System for Modeling and Capacity Planning of Multi-Tier Systems

Although cluster-based multi-tier data centers provide a means for supporting scalable web applications, their ad-hoc configuration poses significant challenges to the performance and economical costs of multi-tier applications. This paper presents the design and implementation of MoKaa utilityaware framework for modeling multi-tier data centers and planning their capacity and optimal configuration. The contribution of the paper is threefold. First, we identify two levels of configuration of cluster-based multi-tier data centers, local configuration that applies at server’s level and architectural configuration that relates to the clusters of servers in a multi-tier architecture. The combination of these two levels of configuration improves the overall performance and cost of cluster-based multi-tier data centers. Second, we present a utility function for characterizing the impact of local and architectural configurations on the performance and cost of multi-tier systems. Third, we develop a utilityaware capacity planning algorithm for efficiently calculating the optimal local and architecural configuration of multi-tier data centers to provide guarantees on performance while minimizing the cost. Our experiments on a multi-tier e-commerce auction site show the effectiveness of MoKa. Moreover, the experiments show that the combination of local and architectural configurations provides a 100% accurate utility for the multi-tier system, while with a single level of optimization (local or architectural) accuracy is limited between 20% and 90%. Key-words: Multi-tier systems, Data centers, Modeling, Capacity planning, QoS, Optimization. ∗ INRIA † Univ. Grenoble I – INRIA in ria -0 03 40 77 2, v er si on 1 25 N ov 2 00 8 2 Arnaud & Bouchenak