Towards Easy and Realistic Network Infrastructure Testing for Large-scale Machine Learning
This addresses the challenge of costly and complex network testing for ML practitioners, though it is incremental as it builds on existing simulation methods.
The paper tackles the problem of testing network infrastructure for large-scale machine learning by introducing Genie, a framework that emulates GPU-to-GPU communication using CPU-initiated traffic and integrates with the ASTRA-sim simulator to model network-ML workload interactions, enabling realistic performance assessment without expensive GPUs.
This paper lays the foundation for Genie, a testing framework that captures the impact of real hardware network behavior on ML workload performance, without requiring expensive GPUs. Genie uses CPU-initiated traffic over a hardware testbed to emulate GPU to GPU communication, and adapts the ASTRA-sim simulator to model interaction between the network and the ML workload.