Enabling Silent Telemetry Data Transmission with InvisiFlow
summary
InvisiFlow routes low-priority telemetry off the shortest path by probing neighbors with pull requests and forwarding along a congestion gradient computed from neighbor egress buffer usage plus distance to collector, using programmable-switch packet generators and recirculation ports to avoid egress drops. Evaluation on FatTree shows near-zero telemetry loss under heavy load (about 33.8× lower loss than Everflow/Planck) while adding negligible user impact versus default-priority (which can raise FCT ~20%) or low-priority (≈10% telemetry loss).
AI-written summary · about
chapters (6)
- 0:00InvisiFlow introduction
- 2:42Telemetry transmission limits
- 4:50InvisiFlow design
- 9:44Architecture and implementation
- 13:14Evaluation results
- 15:40Conclusion and future
related talks
16mSöze: One Network Telemetry Is All You Need for Per-flow Weighted Bandwidth Allocation...
41mNetwork Observability
16mBackpressure Flow Control
17mPrecise Data Center Traffic Engineering with Constrained Hardware Resources
23mVeriFlow: Verifying Network-Wide Invariants in Real Time
24mBlink: Fast Connectivity Recovery Entirely in the Data Plane
25mDebug Functional and Performance Issues in Rte_flow
21mYour Coflow has Many Flows: Sampling them for Fun and Speed