Online automatic energy-saving deployment under QoS guarantee for web server cluster

Abstract

Web server cluster has been widely used to improve the performance and reliability of web servers. Web cluster is usually deployed to handle peak load, resulting in an excessive waste of energy; hence, its deployment should be dynamically adjusted according to real-time load demand. This paper proposes an online automatic energy-efficient deployment scheme for web cluster, which tries to minimize cluster's energy consumption under QoS guarantee. The scheme uses both CPU dynamic frequency scaling and server dynamic switching on/off mechanisms to reduce power consumption. It uses M/G/1 PS queue model to model server, and transforms cluster's deployment problem to a constrained programming problem. Moreover, a hybrid algorithm is proposed to solve the problem. The algorithm adopts greedy idea to determine the on/off state of each server and adopts chaotic optimization to search the optimal solution. Due to the fewer variables and efficient solving algorithm, the problem can be solved online even applied in large-scale clusters. Simulation tests demonstrate that the hybrid solving algorithm can find high-quality solution in less computing time.

A Power and Performance Management Simulation Platform for Web Application Server Cluster

“…We suppose each server node is a real server or the sole VM of a real server [14]. Nowadays, mainstream CPU usually has four or more cores, and each core has many available discrete frequencies, e.g., Intel Xeon E5-1650 has 6 cores and each core has 15 frequencies.…”

“…CloudSim package provides some examples about power-saving simulation of data center, but in these examples, tasks are batch-produced, while actual web request is a kind of streaming load, and energy consumption calculation is not very accurate as they do not consider all possible states of server (such as ignoring the process of server switching on/off). CloudSim has been widely used in various simulation of cloud computing platform, but they all deploy multiple VMs on a Host [13] and save energy by migration and integration of VMs (Virtual Machines), which do not accord with the deployment manner of web cluster [14].…”