Cognitive and Self-Adaptive System for Effective Distributed-Tracing
summary
Claims an span aggregator and a trace clusterer that uses density-based DBSCAN on hundreds of feature vectors to drive self-adaptive sampling, replacing random 1–5% capture with an inverse-frequency sampling rule. System extracts features like origin service, duration, destination and status codes, clusters traces, and raises sampling for low-density (interesting) clusters while lowering it for common ones. Prototype runs online, cuts storage/COGS and surfaces more error/variant traces, and the uploader reports faster MTTR for SRE teams.
AI-written summary · about
chapters (10)
- 0:00Introduction to talk
- 0:35Distributed tracing basics
- 2:03Why use distributed tracing
- 3:28Trace IDs and spans
- 4:53Sampling rate importance
- 5:15Head-based vs tail-based sampling
- 7:03Self-adaptive sampling solution
- 8:50Adaptive sampling architecture
- 10:16Feature vectors for clustering
- 12:07Summary and conclusion
related talks
14mCognitive and SelfAdaptive System for Distributed Tracing
11mCognitive and Self-Adaptive System for Effective Distributed-Tracing...- 14mLightning Talk: Cognitive and SelfAdaptive System for Distributed Tracing
13mDistributed Tracing: Adaptive and Telemetry-Based Approach for Effective...
49mDistributed Tracing: How the Pros Debug Concurrent and Distributed Systems
34mDistributed Tracing, Lessons Learned
26mDistributed Tracing: All the Warning Signs Were Out There!!!
45mTracing: A journey to tactical insights