Harp: Improving VPC Network Availability via Efficient Failure Detection and Rerouting...
summary
Software system that monitors feasible physical underlay paths per VM pair and embeds inband inspection probes into normal packets, then flips the underlay UDP source port to force a different ECMP path when a path is judged bad. Uses SNMP-based path control to build per-pair path pools, runs detection at millisecond cadence, and avoids hardware or transport changes. Deployed in Tencent Cloud for >2 years; production data show sub‑second reroutes, P50 repair <98 ms, and outage time reductions of ~78.7%–99.97%.
AI-written summary · about
chapters (11)
- 0:00Harp and VPC challenges
- 2:46Ideal failure handling requirements
- 4:15Harp solution overview
- 7:28SNMP path control
- 12:10Building path pools
- 12:54In-band failure detection
- 16:12False positives and negatives
- 20:31Harp detection advantages
- 21:14Tencent Cloud performance
- 22:37Comparison with existing
- 23:01VM impact and summary
related talks
29mNetwork Fault Finding System: Packet Loss Triangulation
17mUnlocking ECMP Programmability for Precise Traffic Control
26mF10: A Fault-Tolerant Engineered Network
21mAAsclepius: Monitoring, Diagnosing, and Detouring at the Internet Peering Edge
32mActive Fault Finding in Networks
12mStaying Alive: Connection Path Reselection at the Edge
23mAutomatic Traffic Scheduling for Internet Connectivity Failures
12mBreaking the Transience-Equilibrium Nexus: A New Approach to Datacenter Packet Transport