Self-Recoverable and Highly Available Data Structures for NVRAM-centric Database Systems

新硬件技术和架构的出现，如非易失性RAM（NVRAM）、硬件传输存储器（HTM）、远程直接访问存储器（RDMA）和异构核，正在推动对现有数据库架构进行必要的重新思考，以利用硬件技术进步带来的新机遇。在这个建议中，我们专注于NVRAM对数据库体系结构的影响，特别是对NVRAM的数据库数据结构的基础研究。我们计划设计高度可用的混合数据结构，以缩小持久性内存和易失性内存之间的差距。一方面，我们的目标是使这些数据结构具有崩溃安全性，即，从任何软件或硬件崩溃场景中自恢复，使得它们可以在重启后以最小的恢复工作用于读取和写入操作。另一方面，我们打算设计高度可扩展的并发方案，再加上任何点崩溃恢复和细粒度的复制技术，是必要的设计，以实现前一个目标。将特别强调端到端场景的实验评估和系统集成。

The advent of new hardware technologies and architectures, such as Non-Volatile RAM (NVRAM), Hardware Transactional Memory (HTM), Remote Direct Access Memory (RDMA), and heterogeneous cores, is driving a necessary rethink of existing database architectures to leverage the new opportunities brought by advancement in hardware technology. In this proposal, we focus on the impact of NVRAM on database architectures, in particular on fundamental research on database data structures for NVRAM. We plan to devise highly available hybrid data structures that close the gap between persistent memory and volatile memory. On the one hand, we aim to make these data structures crash-safe, i.e., self-recoverable from any software or hardware crash scenario, such that they can be used for both read and write operations after a restart with minimal recovery effort. On the other hand, we intent to design highly scalable concurrency schemes that are coupled with the any-point crash recovery and fine-grained replication techniques that are necessary to devise in order to achieve the former goal. Special emphasis will be placed on experimental evaluation and system integration for end-to-end scenarios.