ZooRoute: Enhancing Cloud-Scale Network Reliability via Candidate Path Provisioning...
summary
ZooRoute implements tenant‑transparent, underlay‑agnostic failover by changing the VXLAN outer source port and maintaining per‑destination candidate port sets probed proactively; it uses a bitmap and hierarchical indexing to compress Tofino state (about 15× compression), packet‑triggered register updates and loopback reinjection to avoid PCIe bottlenecks (about 1000× faster updates), and on‑host adaptive probing to cut CPU cost by >95%. Deployed 26 months at Alibaba Cloud, it achieves one-shot deterministic reroute, reduces cumulative outage time by 93.19% and masks 98.21% of failures, typically recovering within 1–3 seconds.
AI-written summary · about
chapters (9)
- 0:00ZooRoute introduction and goals
- 1:22Network failure recovery challenges
- 2:46Candidate path provisioning idea
- 3:50ZooRoute system architecture
- 4:35Scaling challenges and solutions
- 7:01Stateful middle boxes handling
- 8:24Production results effectiveness
- 9:49Limitations and lessons
- 10:52Summary and future directions
related talks
16mOptimizing Network Provisioning through Cooperation
15mZeta: A Scalable and Robust East-West Communication Framework in Large-Scale Clouds
33mCheap Routes, Fast Paths & Keeping Track
12mAccessing Cloud with Disaggregated Software-Defined Router
26mF10: A Fault-Tolerant Engineered Network
23mAutomatic Traffic Scheduling for Internet Connectivity Failures
24mBlink: Fast Connectivity Recovery Entirely in the Data Plane
18mTGW: Operating an Efficient and Resilient Cloud Gateway at Scale
