Revolutionizing Datacenter Networks via Reconfigurable Topologies | Stefan Schmid | TU Berlin скачать в хорошем качестве

Revolutionizing Datacenter Networks via Reconfigurable Topologies | Stefan Schmid | TU Berlin

Follow SAI Conferences on Linkedin:   / saiconference   Conference Website: https://saiconference.com/FICC Stefan Schmid is a Full Professor at TU Berlin and part-time researcher at Fraunhofer SIT. A Principal Investigator at the Weizenbaum Institute, his research spans distributed and networked systems. He earned his MSc and PhD from ETH Zurich and has held research positions across Europe, including at Aalborg University, University of Vienna, CNRS, and UCL. In this engaging and forward-looking keynote titled "Revolutionizing Datacenter Networks via Reconfigurable Topologies," Professor Stefan Schmid explores how emerging optical switching technologies are enabling a shift from fixed, demand-oblivious datacenter architectures to dynamic, demand-aware ones. With the explosive rise of cloud computing and data-intensive applications such as machine learning, today’s datacenter networks are under growing pressure. This talk delves into how Reconfigurable Datacenter Networks (RDCNs) can unlock a new level of efficiency by dynamically adapting to traffic patterns—much like changing highway lanes based on congestion. Professor Schmid introduces a rich taxonomy for classifying datacenter networks, discusses technological enablers such as tunable lasers and programmable mirrors, and presents a complexity map showing how different workloads—from batch processing to deep learning—exhibit varying spatial and temporal structures. He explains how leveraging this structure through demand-aware and self-adjusting topologies can significantly improve performance, bandwidth utilization, and energy efficiency. Drawing from his Communications of the ACM article and supported by real-world case studies like Google's Jupiter network, this keynote offers both high-level vision and deep technical insight. 📌 Key Topics Covered: Limitations of fixed datacenter topologies Motivation and benefits of reconfigurable architectures Technological foundations: photonics, optical circuit switching Taxonomy: static vs. dynamic, demand-oblivious vs. demand-aware Real-world applications and workload-specific strategies Challenges in routing, congestion control, and system design