MAC: Metadata Acceleration for Sustainable Performance in Big-Data Systems with CXL DRAM
Talkat 20th USENIX Symposium on Operating Systems Design and Implementation (OSDI '26) · 2026-07-13..07-15·
summary
MAC targets Linux metadata spillover into CXL DRAM and places metadata in CXL DRAM while offloads page reclamation to NMP accelerators. With a 1:4 DDR:CXL ratio metadata can exceed 20% of DDR, causing background reclaim to slow and application tail latency to grow over 10×; MAC’s FPGA prototype and two‑NUMA-node emulation show a 30% faster reclamation pipeline and an 82% tail‑latency drop on Neo4j (uploader claims up to reduces tail latency by up to 98%).
AI-written summary · about
chapters (7)
- 0:00MAC and CXL DRAM intro
- 2:59Metadata spillover tail latency
- 4:47Page reclamation slows in CXL DRAM
- 7:15Fast memory reclamation needs
- 8:40MAC design challenges
- 13:00Experimental results gains
- 14:45Summary
related talks
15mImproving Performance of Flash Based Key-Value Stores Using Storage Class Memory...
16mJohnny Cache: the End of DRAM Cache Conflicts (in Tiered Main Memory Systems)
13mBMC: Accelerating Memcached using Safe In-kernel Caching and Pre-stack Processing
43mPromise and Pitfalls of Persistent Memory
17mPreview: Persistent Memory
17mBreakthrough Data-Centric Computing with a New Memory Tier
28mAchieving 10Gbps Line-rate Key-value Stores with FPGAs
17mHardware Accelerated Big Data Processing