华为云存储创新LAB实习生论文被OSDI’23接收！提出并论证Radix Tree是最适合分离式内存架构的范围索引
OSDI (USENIX Symposium on Operating Systems Design and Implementation)是计算机系统领域最顶级的会议，与SOSP齐名，本届会议共收到投稿255篇，录用54篇，录用率21%。华为云存储创新Lab提出并论证Radix Tree是最适合分离式内存架构的范围索引，并设计了第一个面向分离式内存架构的Radix Tree索引SMART。这是继RACE (USENIX ATC’21)、FORD (FAST’22)、FUSEE (FAST’2023)、ROLEX (FAST’23)之后，我们在分离式内存系统方向上的又一个重要研究成果。
【论文信息】
Xuchuan Luo, Pengfei Zuo, Jiacheng Shen, Jiazhen Gu, Xin Wang, Michael Lyu, Yangfan Zhou, “SMART: A High-Performance Apative Radix Tree for Disaggregated Memory”, accepted and to appear in the 17th USENIX Symposium on Operating Systems Design and Implementation (OSDI), 2023.
【论文简介】
内存分离架构将计算资源和内存资源解耦，使两者通过高速网络互联，以获得高的资源利用率和灵活的弹性。现有基于内存分离架构的范围索引都是基于B+树的，然而B+树固有的读写放大问题会消耗大量的网络带宽，导致低吞吐量和高访问时延。在这篇论文中，我们提出基数树才是更加适合内存分离架构的范围索引。这是因为相比于B+树，基数树的读写放大问题会小很多。然而，在内存分离架构下构建一个基数树索引面临着三方面的挑战：昂贵的基于锁的并发控制、有限的内存池端IOPS和复杂的计算端缓存一致性问题。为了解决这些挑战，我们设计了SMART，第一个高性能的基于内存分离架构的基数树索引，并提出了一种混合式的并发控制方案来减小昂贵的锁开销、一种计算端的读代理和写合并技术来减小冗余的网络I/O、一种简单有效的反向校验机制来处理基数树中计算端缓存的一致性问题。实验结果显示，相比于现有最优的内存分离架构上的B+树方案，SMART在写密集负载上提升了6.1倍的吞吐量，在只读负载上提升了2.8倍的吞吐量。该论文是由来自复旦大学的实习生罗旭川在云存储创新LAB左鹏飞博士的指导下完成。
附：华为云存储创新LAB在分离式内存方向上的研究成果
1.Pengfei Zuo, Jiazhao Sun, Liu Yang, Shuangwu Zhang, Yu Hua, “One-sided RDMA-Conscious Extendible Hashing for Disaggregated Memory”, Proceedings of the USENIX Annual Technical Conference (USENIX ATC), 2021.
2.Ming Zhang, Yu Hua, Pengfei Zuo, Lurong Liu, “FORD: Fast One-sided RDMA-based Distributed Transactions for Disaggregated Persistent Memory”, Proceedings of the 20th USENIX Conference on File and Storage Technologies (FAST), 2022.
3.Pengfei Zuo*, Qihui Zhou*, Jiazhao Sun, Liu Yang, Shuangwu Zhang, Yu Hua, James Cheng, Rongfeng He, Huabing Yan, “RACE: One-Sided RDMA-Conscious Extendible Hashing”, ACM Transactions on Storage (TOS) 18.2 (2022): 1-29.
4.Jiacheng Shen, Pengfei Zuo, Xuchuan Luo, Tianyi Yang, Yuxin Su, Yangfan Zhou, Michael Lyu, “FUSES: A Fully Memory-Disaggregated Key-Value Store”, accepted and to appear in the 21st USENIX Conference on File and Storage Technologies (FAST), 2023.
5.Pengfei Li, Yu Hua, Pengfei Zuo, Zhangyu Chen, Jiajie Sheng, “ROLEX: A Scalable RDMA-oriented Learned Key-Value Store for Disaggregated Memory Systems”, accepted and to appear in the 21st USENIX Conference on File and Storage Technologies (FAST), 2023.
6.Xuchuan Luo, Pengfei Zuo, Jiacheng Shen, Jiazhen Gu, Xin Wang, Michael Lyu, Yangfan Zhou, “SMART: A High-Performance Apative Radix Tree for Disaggregated Memory”, accepted and to appear in the 17th USENIX Symposium on Operating Systems Design and Implementation (OSDI), 2023.
