Analysis and Characterization of Cloud Based Data Center Architectures for Performance, Robustness, Energy Efficiency, and Thermal Uniformity

Abstract

Cloud computing is anticipated to revolutionize the Information and Communication Technology (ICT) sector and has been a mainstream of research over the last decade. Today, the contemporary society relies more than ever on the Internet and cloud computing. However, the advent and enormous adoption of cloud computing paradigm in various domains of human life also brings numerous challenges to cloud providers and research community. Data Centers (DCs) constitute the structural and operational foundations of cloud computing platforms. The legacy DC architectures are inadequate to accommodate the enormous adoption and increasing resource demands of cloud computing. The scalability, high cross-section bandwidth, Quality of Service (QoS) guarantees, privacy, and Service Level Agreement (SLA) assurance are some of the major challenges faced by today’s cloud DC architectures. Similarly, reliability and robustness are among the mandatory features of cloud paradigm to handle the workload perturbations, hardware failures, and intentional attacks. The concerns about the environmental impacts, energy demands, and electricity costs of cloud DCs are intensifying. Energy efficiency is one of mandatory features of today’s DCs. Considering the paramount importance of characterization and performance analysis of the cloud based DCs, we analyze the robustness and performance of the state-of-the-art DC architectures and highlight the advantages and drawbacks of such DC architecture. Moreover, we highlight the potentials and techniques that can be used to achieve energy efficiency and propose an energy efficient DC scheduling strategy based on a real DC workload analysis. Thermal uniformity within the DC also brings energy savings. Therefore, we propose thermal-aware scheduling policies to deliver the thermal uniformity within the DC to ensure the hardware reliability, elimination of hot spots, and reduction in power consumed by cooling infrastructure. One of the salient contributions of our work is to deliver the handy and adaptable experimentation tools and simulators for the research community. We develop two discrete event simulators for the DC research community: (a) for the detailed DC network analysis under various configurations, network loads, and traffic patterns, and (b) a cloud scheduler to analyze and compare various scheduling strategies and their thermal impact.