Given the unabated growth of storage in data centers, its energy footprint continues to grow, which makes storage energy management a crucial issue. Furthermore, with the emerging trend of underprovisioning of power and cooling infrastructures in large facilities, it is important to flexibly adapt the entire infrastructure of a client, including its storage system, to changing energy limitations. In this paper, we presentan energy-adaptive framework called flexStore that provides a flexible mechanism to specify and control the energy vs. performance tradeoffs. These mechanisms are defined and enforced by a software layer called “Policy Engine” that controls the number of active copies of deduplicated data chunks in storage containers based on energy availability. The mechanism synchronizes data chunks in inactive storage containers with those in the active containers, in the background to enable them to be put into service quickly when needed. We evaluate flexStore with different workloads on a sample data center environment and demonstrate the effectiveness of its control mechanisms in adapting to the performance and energy constraints.
Bibliographical noteFunding Information:
This research was supported by NSF, USA grant CNS-1422921 and CNS-1421913.
© 2019 Elsevier B.V.
- Adaptive deduplication
- Energy adaptive storage
- Flexible distributed storage
- Software defined energy management