Architecture Support for Crash Consistency in Persistent Memory

Jungi Jeong, Purdue University
Computing Abstraction

Abstract: Non-volatile memory (NVM) technologies are closing the performance gap between storage and memory. However, the integrity of persistent data structures after an unclean shutdown remains a concern.

This talk highlights architectural support that provides atomicity and ordering of NVM writes. For atomicity, I revisited traditional write-ahead logging (WAL) techniques in database systems. However, since NVM shows access latency comparable to DRAM, past wisdom on logging does not hold. In this talk, I will present a novel ReDU architecture that is optimized for redo-logging in persistent memory.

Second, for ordering, I will present PMEM-Spec that speculatively allows PM accesses without stalling or buffering, and detects ordering violation (i.e., misspeculation). PMEM-Spec treats misspeculation as power failure and thus leverages failure-atomic transactions to recover from misspeculation by aborting and restarting them purposely. Since the ordering violation rarely occurs, PMEM-Spec can accelerate persistent memory accesses without a significant misspeculation penalty.

Meeting ID: 367696905 / Participant passcode: 3297