UC San Diego
Providing Easy to Use and Fast Programming Support for Non-Volatile Memories
- Author(s): Memaripour, Amirsaman
- Advisor(s): Swanson, Steven
- et al.
Non-Volatile Memory (NVM) technologies, such as 3D XPoint, offer DRAM-like performance and byte-addressable access to persistent data. NVMs promise an opportunity for fast, persistent data structures, and a wide range of applications stand to benefit from the performance potential of these technologies. These potential benefits are greatest when applications access NVM directly via load/store instructions rather than conventional file-based interfaces. Directly accessing NVM presents several challenges. In particular, applications need guaranteed consistency and safety semantics to protect their data structures in the face of system failures and programming errors.
Implementing data structures that meet these requirements is challenging and error-prone. Existing methods for building persistent data structures require either in-depth code changes to an existing data structure or rewriting the data structure from scratch. Unfortunately, both of these methods are labor-intensive and error-prone.
Failure-atomicity libraries and programming language extensions can simplify this task. However, all the proposed solutions either require pervasive changes to existing software or incur unacceptable overheads to runtime performance. As a result, porting legacy applications to leverage NVM is likely to be prohibitively difficult and time-consuming.
This dissertation first presents Breeze, an NVM toolchain that minimizes the changes necessary to enable legacy code to reap the benefits of directly accessing NVM. In contrast to PMDK and NVM-Direct, Breeze reduces the programming effort of porting Memcached and MongoDB by up to 2.8×, while providing equal or superior performance.
Second, it introduces NVHooks, a compiler that automatically annotates NVM accesses and avoids disruptive and error-prone changes to programs. NVHooks reduces the cost of these annotations by applying novel, NVM-specific optimizations to their placement. For our tested benchmarks, NVHooks matches the performance of hand-annotated code while minimizing programmer effort.
Finally, it presents Pronto, a new NVM library that reduces the programming effort required to add persistence to volatile data structures. Pronto uses asynchronous semantic logging (ASL) to allow adding persistence to the existing volatile data structure (e.g., C++ Standard Template Library containers) with minor programming effort. ASL moves most durability code off the critical path. Our evaluation shows Pronto data structures outperform highly-optimized NVM data structures by a large margin.