From Ideal to Practice: Data Encryption in eADR-based Secure Non-Volatile Memory Systems

TL;DR

This paper introduces cost-efficient encryption schemes for eADR-based NVM systems that ensure data confidentiality without significantly impacting performance, bridging the gap between persistence and security.

Contribution

It proposes novel encryption schemes, BBE and Sepencr, tailored for ideal and practical eADR execution models to enhance data confidentiality efficiently.

Findings

BBE supports encryption via eADR battery during crashes.

Sepencr encrypts cached data at system startup using OTPs.

Significant performance overhead reduction compared to in-cache encryption.

Abstract

Extended Asynchronous DRAM Refresh (eADR) proposed by Intel extends the persistence domain from the Non-Volatile Memory (NVM) to CPU caches and offers the persistence guarantee. Due to allowing lazy persistence and decreasing the amounts of instructions, eADR-based NVM systems significantly improve performance. Existing designs however fail to provide efficient encryption schemes to ensure data confidentiality in eADR-based NVM systems. It is challenging to guarantee both data persistence and confidentiality in a cost-efficient manner due to the transient persistence property of caches in eADR. Once the system crashes, eADR flushes the unencrypted data from the cache into NVM, in which security issues occur due to no encryption. To bridge the gap between persistence and confidentiality, we propose cost-efficient BBE and Sepencr encryption schemes that efficiently match different eADR execution models from ideal to practice. Under the ideal eADR execution model, BBE supports the encryption module via the battery of eADR upon crashes. Under the practical eADR execution model, Sepencr generates the one-time paddings (OTPs) at the system startup to encrypt the cached data in case the system crashes. Our evaluation results show that compared with an intuitive in-cache encryption scheme in eADR-based systems, our designs significantly reduce performance overheads while efficiently ensuring data confidentiality.