SIMON: A Simple and Scalable Method for Sensing, Inference and Measurement in...
summary
SIMON reconstructs per-switch queue averages from edge TX/RX timestamps and a small set of probes, solves an underdetermined linear system with Lasso and then replaces it with a trained neural net for near real-time output. Key specifics: probes at ~1% bandwidth, 5,000–10,000× speedup over iterative Lasso, measured accuracy of 3–5 KB queue error and link utilization <1%, validated with NetFPGAs, cross-validation, and deployments on 1/10/40 G testbeds; requires path knowledge and microsecond-class clock sync.
AI-written summary · about
chapters (9)
- 0:00Start
- 0:27Motivation and challenges
- 0:57Current measurement methods
- 2:58Reconstruct Average Queue Length in a Recon-interval
- 5:31Power Spectral Density of the Queue Process Cont'
- 8:09Network tomography
- 13:21Speeding Up Reconstruction with Neural Networks
- 14:16RMSE and Speedup
- 15:51Inference of the Journey of A Data Packet
related talks
46mSelf-Programming Networks: Network Clock Synchronization, Event Playback and Real-time Action
26mExploiting a Natural Network Effect for Scalable, Fine-grained Clock Synchronization
22mSoftware Defined Traffic Measurement with OpenSketch
12mToward Nearly-Zero-Error Sketching via Compressive Sensing
16mHeteroSketch: Coordinating Network-wide Monitoring in Heterogeneous and Dynamic Networks
22mLancet: A self-correcting Latency Measuring Tool
21mMillisampler: Fine-Grained Network Traffic Analysis
21mQ&A Session 1 – Neil Spring