Scalable Crash Consistency for Secure Persistent Memory
TimeWednesday, July 13th4:10pm - 4:30pm PDT
Location3004, Level 3
Embedded Memory, Storage and Networking
DescriptionPersistent memories suffer from security and crash consistency issues. Bonsai-Merkle-Tree (BMT) uses metadata to ensure security. Data and its metadata need to be atomically persisted for recovery. To guarantee crash consistency, transactions are widely used but the BMT update increases the latency and write traffic. This paper presents Secon to efficiently ensure SEcurity and crash CONsistency by designing scalable write-through metadata caches. Secon leverages a transaction-specific epoch persistency model to accelerate transactions. To reduce writes, Secon co-locates counters with logs and coalesces BMT blocks. Experimental results demonstrate that Secon reduces the transaction latency and write traffic over the state-of-the-art design.