Abstract

The emerging generation of cloud computing systems demand high-performance, energy efficiency, and dense, high-bandwidth memory. An emerging concern is in-memory encryption to protect data in these environments. New memory technologies, such as phase-change memory (PCM), have been proposed to improve storage density and energy efficiency, with a near DRAM performance, while suffering from endurance challenges. Industry leaders like Intel and Microsoft have established AES-XTS encryption as the de facto standard to provide the required information security in nonvolatile memories like PCM. The energy and endurance challenges of PCM are exacerbated when AES-XTS encryption is applied. To address these challenges, we present MACE, a technique to improve the energy-efficiency and lifetime of PCM in a server environment using AESXTS encryption. Additionally, we propose an architecture for MACE called WINDU which leverages in-memory, lightweight compression to reduce the auxiliary storage overhead of MACE, securing up to a 2.6 $\times$ lifetime improvement beyond existing techniques with the same area overhead. We also propose LARS, a new cradle-to-grave energy evaluation framework that considers system lifetime. Using LARS, we provide a holistic analysis of MACE and WINDU, demonstrating how the 15% mean reduction in dynamic energy and lifetime improvements inform system design and memory selection.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call