PhD in Communication-Optimized Inference in Large Language Models

Eindhoven University of Technology offers a fully funded PhD position focusing on communication-optimized inference for large language models within the COIL project. The successful candidate will combine mathematical modeling, such as Network Calculus, with experiments on programmable network hardware to optimize GPU interconnects. Working across an interdisciplinary team, the researcher will spend six months collaborating with the Technical University of Munich. The candidate is expected to develop analytical models, build prototypes, conduct reproducible testbed experiments, and disseminate findings openly.

• Develop fast, composable analytical models using Network Calculus to capture network traffic and latency bounds during LLM inference.
• Investigate how architectural choices like network topology and parallelization strategies impact performance and energy consumption.
• Design runtime decision tools for workload placement, routing, and prioritization within AI clusters.
• Perform experimental validation using programmable networking testbeds including P4-programmable switches and SmartNICs.
• Collaborate with partner organizations such as SURF to evaluate experimental results on real AI infrastructure.
• Spend a mandatory six-month research period at the Technical University of Munich (TUM) working with the Network Architectures and Services group.
• Release research tools openly and share findings with networking standardization communities.