Intern/Master Student — DC-OXC Traffic Scheduling for Super-Node Architecture

Huawei; München; Vor Ort; EUR 13.000 - 20.000; Vollzeit

Benefits dieser Stelle

Healthy meals in canteen

Training opportunities

Language courses German/Mandarin

Global team environment

Zusammenfassung

Huawei Heisenberg Research Center (Munich) seeks an Intern/Master Thesis Student for DC-OXC Traffic Scheduling in a Super-Node Architecture. You will explore MoE traffic, prototype load‑aware routing, and aim for deterministic traffic using OCS interconnects.

Strong concept familiarity with MoE, DL basics, and optimization is required. You will collaborate in a research‑oriented environment with opportunities for training and hands‑on development on cutting‑edge network tech.

Qualifikationen

• Pursuing a Bachelor's or Master's degree in a related field.
• Solid knowledge of neural networks, training/inference pipelines and GPUs.
• Strong background in graph theory and optimization (linear/integer).
• Familiarity with multi-commodity flow or convex optimization is a plus.
• Understanding of MoE models and token routing concepts is expected.
• Proficient in written and spoken English.

Aufgaben

• Study MoE traffic scheduling and load imbalance impacts on Alltoallv congestion.
• Research techniques to make MoE traffic deterministic for OCS.
• Investigate strategies to avoid frequent topology changes using optical bandwidth.
• Design load‑aware routing to distribute non‑uniform Token traffic.
• Develop software-level traffic smoothing to reduce congestion jitter.

Kenntnisse

Deep learning fundamentals

Graph theory

Optimization (linear/integer)

MoE models understanding

Network interconnects

English fluency

Ausbildung

Bachelor's or Master's in CS/CE/Applied Math

Jobbeschreibung

Huawei Heisenberg Research Center (Munich) is responsible for advanced technology research, architectural development, design and strategic engineering of our products.

The Optical Technology Laboratory in Munich researches and develops signal processing, components, systems, and optical switching for fixed access, data center networks, metro and long-haul optical networks. The department engages in local research and collaborates with academic and industrial partners from the European ecosystem.

Join us as a

Intern/Master Thesis Student — DC-OXC Traffic Scheduling for Super-Node Architecture (m/f/d)

Your mission

• Study existing approaches to MoE traffic scheduling, collective communication optimization, and OCS-based interconnect architectures, build a clear understanding of how load imbalance in MoE inference leads to Alltoallv congestion and long-tail latency.
• Research and implement techniques to transform random, dynamic sparse MoE traffic into deterministic, uniform traffic patterns that OCS can efficiently handle.
• Investigate collective communication strategies that eliminate the need for frequent topology reconfiguration, leveraging optical path bandwidth to remove local hotspots and long-tail effects
• Design and prototype load-aware routing algorithms that intelligently distribute non-uniform Token traffic across limited OCS "narrow pipes.
• Develop software-level traffic smoothing mechanisms to eliminate physical-layer congestion jitter, targeting end-to-end inference latency equivalent to 100% load-balanced scenarios

Your areas of expertise

• Currently pursuing a Bachelor's, Master's degree in Computer Science, Computer Engineering, Electronic Engineering, Applied Mathematics, or a related field.
• Solid understanding of deep learning fundamentals, including neural network architectures, training/inference pipelines, and GPU computation.
• Strong background in graph theory, combinatorial optimization, linear/integer programming, or queuing theory.
• Familiarity with multi-commodity flow optimization, network flow problems, or convex optimization is a strong plus.
• Understanding of network interconnect technologies, including OCS (Optical Circuit Switching), InfiniBand, RoCE, or similar high-performance interconnects is a strong plus.
• Basic understanding of MoE (Mixture of Experts) models — how Token routing works and why it creates non-uniform communication patterns. Deep expertise is not expected, but conceptual familiarity is required
• Ability to work independently and collaboratively in a research-oriented environment
• Fluent in written and spoken English language

Your rewards of working here

• Our culture is characterized by innovative power and team spirit as well as the intensive exchange of knowledge and experience within our global network.
• We offer healthy meals ranging from traditional Chinese to western delicacies in our famous company canteen.
• To keep your development ongoing, you will find a broad range of training opportunities. Many online and face-to-face training programs incl. language courses in German and Mandarin.
• Our diverse and welcoming environment is shaped by different backgrounds and around 40 individual nationalities.
• Self-responsible work in a competent, motivated and constantly growing team.

Huawei is a leading global information and communications technology (ICT) solutions provider. Our ICT solutions, products and services are used in more than 170 countries and regions, serving over one-third of the world's population. With 208,000 employees, Huawei is committed to develop the future information society and build a Better Connected World.