tpprof: A Network Traffic Pattern Profiler
summary
tpprof profiles running networks by extracting network states (representative switch-level utilization sampled via SNMP/sFlow) and traffic pattern subsequences (finite-state signatures). It clusters samples with Gaussian mixture models after PCA, implements fuzzy matching of state-machine signatures to raise alerts, and runs in Python/C++. Evaluation uses a 4-switch leaf‑spine testbed with 6 servers (memcached workloads) and a Facebook front-end trace; it discovers a rare state occurring 3.8% of the time and successfully detects all‑to‑all, straggler, and ECMP imbalance patterns.
AI-written summary · about
chapters (7)
- 0:00Introduction and motivation
- 2:40TB Prof system overview
- 4:45Workflow and methodology
- 8:05Design challenges
- 9:20System implementation
- 13:35Evaluation and case studies
- 17:15Conclusion and Q&A
related talks
32mActive Fault Finding in Networks
29mFinding the Needle in a Haystack: Tenant-Level Network...
17mReasoning about Network Traffic Load Property at Production Scale
46mLessons learned from monitoring a large production network
24mPythia: An Automated, Cross-layer Instrumentation Framework for Diagnosing Performanc...
16mClosed-loop Network Performance Monitoring and Diagnosis with SpiderMon
29mNetwork Fault Finding System: Packet Loss Triangulation
24mDeep Network Traffic Observability with Pktvisor and Prometheus