Concordia: Distributed Shared Memory with In-Network Cache Coherence
summary
Concordia implements a DSM that pushes cache coherence into programmable switches, using a hybrid protocol FLOWCC that stores per-block metadata and a read‑writer log to serialize conflicting requests in one network round. It adds ownership migration to keep switch state under the ~10 MB hardware limit and idempotent operations to tolerate packet loss. The authors build a key‑value store, graph engine and port a transaction system and report up to 4.2x, 2.3x and 2x speedups over prior DSMs.
AI-written summary · about
chapters (14)
- 0:00Intro
- 1:11Caching & Coherence
- 2:09Opportunities from Programmable Switches
- 2:59Challenges
- 4:06Concordia Architecture
- 4:51Key Designs
- 5:34In-Network Coherence Protocol
- 7:08Ownership Migration
- 7:48Idempotent Operations
- 8:33More details: checkout our paper
- 8:43Outline
- 9:18Micro Benchmarks
- 10:23Applications
- 11:16Summary & Conclusion
related talks
18mCCC: Re-architecting Delay-based Congestion Control in Datacenter Networks
17mDuhu: Shared Disaggregated Memory for Distributed Data Processing Frameworks
25mDistCache: Provable Load Balancing for Large-Scale Storage Systems with Distributed...
20mAn Automated Scalable RISC-V Cache Coherency Verification Project
21mDistCache: Provable Load Balancing for Large-Scale Storage Systems with Distributed Caching
22mAn Open and Coherent Memory Centric Architecture Enabled by RISC V
19mFarReach: Write-back Caching in Programmable Switches
31mUnshackling Memory!