UC San Diego

Persistent memory (PMEM) allows direct access to persistent storage via a load/store interface. It promises to realize a vision of high performance, data persistence, a simple programming interface, low cost with minimal storage overhead. Previously, processor caches backed by PMEM were volatile, complicating the design of persistent applications and reducing their performance. The new generation of systems with flush-on-fail semantics provides persistent caches, offering the potential for much simpler, faster PMEM programming and execution models.

PMEM programming systems provide the means to apply sets of writes to persistent states atomically. Unfortunately, most of these systems impose significant overhead and are not easy to use. Moreover, most existing approaches to incorporating PMEM realize only a subset of the benefits of PMEM.

This dissertation first presents Clobber-NVM, a failure-atomicity library that ensures data consistency by reexecution. Clobber-NVM’s novel logging strategy, clobber logging, records only those transaction inputs that are overwritten during transaction execution. Then, after a failure, it recovers to a consistent state by restoring overwritten inputs and reexecuting any interrupted transactions. Clobber-NVM utilizes a clobber logging compiler pass for identifying the minimal set of writes that need to be logged. Based on our experiments, Clobber-NVM provides up to 2.5× performance improvement over existing solutions.

Second, it introduces Whole Process Persistence (WPP), a new programming model for systems with persistent caches. In the WPP model, all process state is made persistent. On restart after a power failure, this state is reloaded and execution resumes in an application-defined interrupt handler. We describe the Zhuque runtime, which transparently provides WPP by interposing on the C bindings for system calls in userspace. It requires little or no programmer effort to run applications on Zhuque. Our measurements show that Zhuque significantly outperforms state-of-the-art PMEM libraries. More important, unlike existing systems, Zhuque places no restrictions on how applications implement concurrency, allowing us to run a newer version of Memcached on Zhuque and gain more than 7.5× throughput over the fastest existing persistent implementations.

Finally, it presents PERSISTRON , a key-value store that exploits PMEM effectively with almost no change to the storage layer. Our main observation is that most key-value stores employ a cache to avoid expensive access to storage. By moving the cache layer to PMEM and using regular cache-management functions to manage PMEM, most code is untouched, and the cache is potentially reusable for other applications. We have implemented PERSISTRON by modifying SplinterDB. PERSISTRON can improve query performance by up to 46% over a cost-equivalent all-DRAM configuration of SplinterDB.