Crystalor: Persistent Memory Encryption Mechanism with Optimized Metadata Structure and Fast Crash Recovery

This study presents an efficient persistent memory encryption mechanism, named Crystalor, which efficiently realizes a secure persistent/non-volatile memory based on an authentication tree with structural optimization, such as the split counter (SC). Crystalor can completely exploit the advantage of metadata compression techniques, whereas existing mechanisms are incompatible with such optimization. Meanwhile, Crystalor incurs almost no latency overhead under the nominal operation conditions for realizing the crash consistency/recoverability. We implement Crystalor with a state-of-the-art parallelizable authentication tree instance, namely ELM (IEEE TIFS 2022), and evaluate the effectiveness by both algorithmic analyses and system-level simulation in comparison with the existing state-of-the-art ones (e.g., SCUE in HPCA 2023). For protecting a 4 TB memory, Crystalor requires 29–62% fewer clock cycles per memory read/write operation than SCUE owing to the compatibility with the SC. In addition, Crystalor and SCUE require 312GB and 554GB memory overheads for metadata, respectively, which indicates that Crystalor achieves a reduction of memory overhead by 44%. The result of the system-level simulation using the gem5 simulator indicates that Crystalor achieves a reduction of the workload execution time by up to 11.5% from SCUE. Moreover, Crystalor can offer a lazy recovery, which makes recovery several thousand times faster than SCUE