SEA: SSD Staged Energy Efficient Object Storage System Architecture

• Published in: International journal of innovative technology and exploring engineering Vol. 9; no. 7; pp. 884 - 893
• Main Authors: Padigi, Sai Vishwas, Kulkarni, Sameer, HL, Phalachandra, Sitaram, Dinkar
• Format: Journal Article
• Language: English
• Published: 30.05.2020
• ISSN: 2278-3075; 2278-3075
• DOI: 10.35940/ijitee.G5693.059720

The advent of social media, smart mobile devices and the Internet of Things (IoT) has led to the generation of unstructured data at an astronomical rate, thereby creating an ever-increasing demand for object storage. These object storage systems consume a lot of energy, resulting in increased heat dissipation, greater cooling requirements (which in turn consumes more energy), higher operational costs, and excessive carbon footprint. Although there has been some progress in building energy-efficient disk systems, works on energy-efficient object storage systems are still in the nascent stage. In this paper, we propose SEA: An SSD Staged Energy Efficient Object Storage System Architecture, wherein we introduce a staging layer comprising Solid State Drives (SSDs) on top of the existing object storage system consisting primarily of Hard Disk Drives (HDDs). SSDs not only consume lesser power as compared to HDDs but are also much faster. Leveraging SSDs for staging reduces the number and frequency of requests hitting the object storage system underneath, allowing us to selectively spin down a substantial number of disks without violating any Service Level Agreements driven by Quality of Service requirements while reducing the total disk energy consumption. Given the high-performance characteristics of SSDs, this SSD staging layer significantly enhances the performance of the object storage system as a whole. As a case study, we have modeled this architecture for OpenStack Swift. Our simulation results using a Dropbox-like workload show that, even after factoring in the additional energy consumed by the SSD staging layer, our model was able to reduce the total disk energy consumption by up to 15.235 % and improve performance by up to 29.06 %.