Deploying User-space TCP at Cloud Scale with LUNA
summary
LUNA is a user-space TCP libOS built with DPDK and SR-IOV/flow-steering that runs on a run-to-completion model (batch and inline r2c), a user-space slab/Z-buffer for zero-copy handoff, and netlink sync to coexist with kernel TCP. Deployed since 2018 in Alibaba Cloud storage, it reports up to 3.5× throughput and up to 53% latency reduction in microbenchmarks, 50% higher EBS throughput in production, and 50–68% lower tail latency for table services.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 0:34LUNA Cloud Storage
- 5:32LUNA Architecture
- 10:52Performance Evaluation
- 12:17Lessons Design Choices
- 17:46Conclusion
related talks
18mThe Evolution of High Performance Networks: A Storage Perspective and Practices
01h 07mLibtpa Introduction
31mA Practical, Research‑Backed Introduction To Cloud‑Native Kernel‑Bypass Networking
17mRearchitecting the TCP Stack for I/O-Offloaded Content Delivery
21mRemote TCP Connection Offload and Applications
37mSPDK Based User Space NVMe Over TCP Transport Solution
24mCan a User-Space Protocol Stack Balance High Performance and Generality? Our Answer: Both.
15mMore Than Capacity: Performance-oriented Evolution of Pangu in Alibaba
