Performance Contracts for Software Network Functions
summary
Defines performance contracts that map NF packet classes to human‑readable formulas over PCVs and claims static prediction of per‑packet instruction and memory access counts without running the NF. Implements BOLT: code analysis + symbolic execution plus a library of analyzed stateful data structures, applied across DPDK and NIC driver. Evaluated on four NFs (Maglev-like LB, NAT, LPM router, MAC bridge) with worst conservative prediction gap ≤ 7.6% and shows usable thresholds for rehashing decisions.
AI-written summary · about
chapters (9)
- 0:00Introduction to performance
- 1:53Performance contracts concept
- 2:46MAC learning bridge example
- 4:45Using performance contracts
- 6:51Generating the contracts
- 12:24Building for NF chains
- 13:56Evaluation and results
- 17:04Use case and summary
- 19:48Q&A session
related talks
17mPerformance Interfaces for Network Functions
15mFormal Methods for Network Performance Analysis
16mAutomatic Parallelization of Software Network Functions
01h 25mIntro + Deep Dive BoF: Telecom User Group and Cloud Native Network Functions (CNF) Testbed
30mCloud Native NFV Performance Benchmarks
16mLemonNFV: Consolidating Heterogeneous Network Functions at Line Speed
36mThe black art of network performances
49mNFV - Research Trends and Market Opportunities