F10: A Fault-Tolerant Engineered Network
summary
Implements an A-B fat-tree topology plus three-tier failover: immediate local rerouting, millisecond-scale pushback broadcasts, and a centralized rebalancer; failure detector watches link bit transitions and reroutes the next packet. Builds a Click/Emulab prototype and a packet-level simulator of 24-port 10Gb switches. Recovers in under 1 ms on the testbed, shows <1/7th packet loss versus Portland in traces, and yields a 30% median MapReduce speedup.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 0:22Next Generation Data Centers
- 1:05Portland
- 1:45Problems with Portland
- 3:17F10 Approach
- 4:24Why is recovery slow
- 9:27Local Rerouting
- 12:11Centralized Controller
- 13:03Failure Detector
- 14:45Questions
- 17:22Simulation
- 18:05Enduser Performance
- 19:00Conclusion
related talks
23mBalancing on the Edge: Transport Affinity without Network State
29mNetwork Fault Finding System: Packet Loss Triangulation
15mFiDe: Reliable and Fast Crash Failure Detection to Boost Datacenter Coordination
32mActive Fault Finding in Networks
12mDebugging Transient Faults in Data Centers using Synchronized Network-wide Packet
16mA High-Speed Stateful Packet Processing Approach for Tbps Programmable Switches
22mAdapting TCP for Reconfigurable Datacenter Networks
04mLoad balancing through the years