AutoSketch: Automatic Sketch-Oriented Compiler for Query-driven Network Telemetry
Talkat 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI '24) · 2024-04-16..04-18·
summary
AutoSketch automatically compiles high-level query operators into sketch instances, then searches configurations to meet a user-specified accuracy intent interface (example target: 99% of cases ≥95% recall and precision). It maps operators to sketches at the operator level, decomposes multi-state group-by into per-state sketches, and uses benchmark-based configuration with sampling and an HS-based initializer to pick minimal-memory parameters. A P4 prototype runs on programmable switches and, in evaluation, matches expressiveness while using much less switch memory than classical sketches.
AI-written summary · about
chapters (9)
- 0:00AutoSketch introduction
- 1:59Deploying sketch telemetry challenges
- 3:07Accuracy intent for error control
- 5:12Operator sketch mapping
- 7:00Handling group-by operators
- 9:29Benchmark sketch configuration
- 11:20Evaluation setup results
- 13:27Future extensions
- 14:11Conclusion & Q&A
related talks
17mSketchovsky: Enabling Ensembles of Sketches on Programmable Switches
16mHeteroSketch: Coordinating Network-wide Monitoring in Heterogeneous and Dynamic Networks
15mSketchLib: Enabling Efficient Sketch-based Monitoring on Programmable Switches
15mSketchPipe: Toward Accurate Sketch-based Network Measurement on Multi-Pipeline...
22mSoftware Defined Traffic Measurement with OpenSketch
28mAutomatic Operator Generation for Open AI frameworks: MindSpore/AKG
12mToward Nearly-Zero-Error Sketching via Compressive Sensing
20mSparseAuto: An Auto-Scheduler for Sparse Tensor Computations Using Recursive Loop Nest(…)