Large-Scale Practice of Persistent Memory in Alibaba Cloud
Talkat KubeCon + CloudNativeCon North America 2021 · on Thursday, 2021-10-14· at 11:00· Concourse Hall 152 + Online·
summary
Alibaba Cloud describes a production PMEM stack that puts persistent memory into Kubernetes as volumes and reports real node capacity to the scheduler, with a node resource manager, pmem csi plugin, auto-resizer and scheduler extensions. It supports kmem/quotapath/lvm/direct types, NUMA-aware placement and prebind/reserve hooks, and claims ~90% DRAM performance at ~70% DRAM cost across millions of DB and memory‑preferred apps. Demonstrations show online resize, quotaed ext4 mounts and cluster rollout working in production.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 0:26Agenda
- 0:53Why is persistence memory
- 1:59Necessary of using Pmap
- 3:56Architecture of Pmap
- 6:22Node Resource Manager
- 11:29Scheduling
- 13:17CPU Scheduling
- 15:51Part 1 Introduction
- 16:19Processing Memory Stacks
- 18:06Benefits
- 19:38Demo
- 27:31Personal Memory
- 29:22Future Work
related talks
43mPromise and Pitfalls of Persistent Memory
35mDeveloping Software for Persistent Memory
35mDeveloping Software for Persistent Memory
17mPreview: Persistent Memory
06mEarly experience of Utilizing Persistent Memory for Database... (Work-in-Progress Report)
24mgo-pmem: Native Support for Programming Persistent Memory in Go
31mReimagining the Future: Where the Market is Evolving
21mPolarDB: Alibaba Cloud-Native Relational Database Service