TGW: Operating an Efficient and Resilient Cloud Gateway at Scale
summary
DPDK-based forwarding plane achieves 2.9× single-node forwarding over prior systems; deployment uses load distributors (LDs) with per-flow core pinning for EIP and pipeline dispatch for stateful CBs. State migration moves major connection state and redirects traffic via BGP with an LD proxy to avoid drops, completing between clusters in ≈4 s without packet loss. Multi-level resilience uses active‑active intra‑AZ, active‑standby inter‑AZ, decentralized migration for malicious flows, and telemetry probers while handling tens of Tbps with packet drop 10^-7–10^-4 and years of 100% availability.
AI-written summary · about
chapters (12)
- 0:00TGW cloud gateway intro
- 1:47TGW architecture challenges
- 3:55TGW design principles
- 4:57TGW workflow
- 6:20Forwarding models
- 8:50Failure resilience models
- 10:55Decentralized mitigation
- 12:00Failure detection system
- 13:26TGW performance
- 14:30TGW best practices
- 16:16Service-driven design
- 16:39Conclusion
related talks
15mZeta: A Scalable and Robust East-West Communication Framework in Large-Scale Clouds
15mCost-effective and Reliable Global Internet Peering with Programmable Switches
34mHyperRouter: Lessons Learnt from Building an L4 Load...
21mManaging 10s of thousands in TPS in a cross-cloud with high availability.
19mTackling DC Congestion and Bursts
52mSuggestions to Enhance DPDK to Enable Migration of User Space Networking Applications to DPDK
17mBifrost: Alibaba's Next-Generation VPC Network with High-Performance Multipath...
18mCStar Gateway: Augmenting Public Cloud Infrastructure for Heterogeneous Network...