Long-Haul RDMA

This project investigates long-haul RDMA for geo-distributed machine learning systems. The project includes:

• Characterization, modeling, and verbs-level emulation of long-haul RDMA behavior.
• Evaluation of whether long-haul RDMA can improve geo-distributed federated learning, including simulation and validation on a real-world testbed.

Related publications:

• Can Long-Haul RDMA Benefit Federated Learning?
• When RDMA Goes Long-Haul: Characterization, Modeling, and Verbs-Level Emulation with Implications for Federated Learning