CausalSim - A Causal Framework for Unbiased Trace-Driven Simulation

Explore a groundbreaking conference talk on CausalSim, an innovative causal framework for unbiased trace-driven simulation. Delve into the challenges of current trace-driven simulators and discover how CausalSim addresses the issue of biased real-world traces. Learn about the framework's approach to learning causal models of system dynamics and latent factors, and its novel tensor completion method for sparse observations. Examine the extensive evaluation of CausalSim on both real and synthetic datasets, including its impressive performance on the Puffer video streaming system. Gain insights into how CausalSim significantly improves simulation accuracy and provides markedly different insights about ABR algorithms compared to biased baseline simulators. Understand the implications of this award-winning research for more accurate and reliable trace-driven simulations in various applications.

NSDI '23 - CausalSim: A Causal Framework for Unbiased Trace-Driven Simulation