Improving Data Center Networks: Physical-Layer Programmability

This lecture discusses the challenges in operating data center networks, focusing on failures, service provisioning, and maintenance. It introduces the concept of physical-layer programmability to restructure network models, improve failure recovery, and enhance transmission performance. The speaker presents innovative network topologies like Clos and random graphs, addressing routing challenges and transmission efficiency. The lecture also covers practical implementations, such as ShareBackup and OmniSwitch, to simplify network management and ensure high-quality service. Moreover, it explores the social impact of network reliability during times of increased utilization. Future directions include end-to-end programmability and joint optimization of traffic and network topology.