Causalsim: Trace-Driven Simulation For Network Protocols
summary
CausalSim is a trace-driven counterfactual simulator that learns a causal network model from an initial RCT (PUFFER) to remove policy bias, maps counterfactual prediction to tensor completion with extremely sparse observations, and trains an estimator with an adversarial neural network training loss that enforces distributional invariances from the RCT. Empirically it aligns simulated buffer-occupancy CDFs with withheld-policy traces and finds ABR dynamics are nearly low-rank (≈99% energy in top two components), improving QoS in simulation.
AI-written summary · about
chapters (7)
related talks
19mEnabling AI Network Cross-Layer Design and Operations with Arcadia...
21mFrom Reactive to Proactive: How AI is Redefining Resilience in Modern Networks
12mDebugging Transient Faults in Data Centers using Synchronized Network-wide Packet
17mTowards provably performant congestion control
21mModel-Agnostic and Efficient Exploration of Numerical State Space of Real-World...
21mLearning in situ: a randomized experiment in video streaming
16mScalable Tail Latency Estimation for Data Center Networks
17mSupercharging Packet-level Network Simulation of Large Model Training via Memoization...