Performance Optimization of Cloud Data Centers with a Dynamic Energy‐Efficient Resource Management Scheme

Abstract

As an advanced network calculation mode, cloud computing is becoming more and more popular. However, with the proliferation of large data centers hosting cloud applications, the growth of energy consumption has been explosive. Surveys show that a remarkable part of the large energy consumed in data center results from over-provisioning of the network resource to meet requests during peak demand times. In this paper, we propose a solution to this problem by constructing a dynamic energy-efficient resource management scheme. As a way of saving energy as well as maintaining cloud user’s quality of experience, the scheme presents a multitier cloud architecture by configuring physical machines (PMs) into two pools: a hot (running) pool and a warm (turned on, but in dynamic sleep) pool. Each PM is configured with a resource search engine (RSE) that finds an available virtual machine (VM) for the request, and a synchronous sleep mechanism is introduced to the warm pool. To analyze the end-to-end performance of the cloud system’s service with the proposed scheme, we establish a hybrid queueing system composed of three stochastic submodels by using a matrix-geometric solution. Accordingly, the average latency of requests and the energy-saving rate of the system are derived. Through numerical results, we show the influence of the synchronous sleep mechanism on the system performance. Moreover, from the perspective of economics, we build a system cost function to study the trade-off between different performance measures. An improved Salp Swarm Algorithm (SSA) is presented to minimize the system cost and optimize the sleep parameter.

Arithmetic Study about Efficiency in Network Topologies for Data Centers

“…It is estimated that power distribution consumes around 15% of the total energy consumption, whilst cooling systems use around 45%, thus leaving the remaining approximately 40% to the IT equipment, which is shared between networking equipment (taking between 30% and 50%, depending on the load level) and computing servers (taking the rest of it) [28]. In this sense, it is to be noted that some part of that energy is consumed in over-provisioning of resources to meet requirements during demand at peak times [29]. Hence, it is crucial to undertake an appropriate network design to optimize the overall performance [30].…”