Investigating Techniques for Improving the Energy Proportionality of Distributed Storage Systems

Abstract

As data-intensive applications become more ubiquitous, the energy consumption problem across distributed systems intensifies, prompting further research to improve these systems. Energy demands across the world are consistently growing, especially due to the storage and processing of data in cloud systems, implying an urgent need for efficient and sustainable data centers to reduce the total overall cost of computing needs. Driven by the environmental and economic costs posed by this challenge, researchers have developed methods to achieve energy proportionality in cloud-related systems through smart resource allocation and workload characteristic consideration. Focusing specifically on the software levels of distributed storage systems, this survey reviews, taxonomizes, and compares a series of publications and research on power-aware and energy-efficient storage techniques and systems. Existing power-reduction data management research is categorized into five separate sectors based on their primary energy-aware technique, spanning classification, layout, replication, de-duplication, and write off-loading. Through this review, we attempt to define the research space and synthesize essential design guidelines and takeaways for the development of future power-aware distributed storage systems; alongside, we present fundamental trade-offs that must be tackled to maintain a high quality of service while reducing power consumption. Centering energy proportionality and efficiency as key design goals, along with factoring in the presented guidelines, can inform new energy-efficient storage systems, thus reducing the harmful environmental impact of cloud computing centers.