Proposes a scalable fluid model to analyze credit-based flow control dynamics in HPC networks, suggesting improved computational efficiency.
Key Points
This research aims to develop a scalable fluid model to capture the complex dynamics of credit-based flow control in high-performance computing interconnections.
Developed a fluid model using coupled delay differential equations to simulate backlog accumulation and credit depletion.
Implemented a fourth-order Runge–Kutta method for integration with history interpolation.
Compared the fluid model's performance against the CODES packet-level simulator for single and multi-hop configurations.
The fluid model achieved a 22–98× speedup compared to the CODES simulator.
It accurately reproduced transient throughput in both single-hop and multi-hop network scenarios.
The model effectively addresses transient dynamics including backpressure and head-of-line blocking.