The building blocks of any large-scale distributed system such as an AI cluster are compute, storage, and network. The network infrastructure connects all the components enabling data transfer, communication, and coordination between compute nodes, storage devices, and AI applications. The network is also the slowest component among the three which necessitates innovative architecture, a scalable software stack, and automation tools.
This book covers the fundamentals of networking, explains the nature of the AI workloads that justify the non-negotiable requirements expected of the network, and explains how those requirements are met. We start with data center architecture and the network topologies that are best suited for AI workloads. We also touch upon automation tools and look at real-world examples from Google, Meta, and OCI to see how these ideas play out. Finally, we talk about security and reliability, introduce you to future trends such as quantum computing and quantum networks, and wrap up by showing how AI and networking have a symbiotic relationship.
By the end of this book, you would have a clear understanding of what kind of topology you would need to run the type of AI workloads you intend to run, and the networking stack you need to use, depending on whether you are building a network for AI training or for AI inference.
WHAT YOU WILL LEARN
● Understand the criticality of networking in AI.
● Networking foundational concepts.
● Datacenter topologies and network architectures.
● Optimize throughput and latency for faster model inference.
● Virtual networking stack and concepts.
● Integrate 5G and quantum networking for edge AI.
● Implement SDN/NFV and RDMA protocols for flexible control.
● Insights into hyperscalar data center network internals.
● Future trends in AI networking.
WHO THIS BOOK IS FOR
This book is for network architects, cloud networking engineers, AI infrastructure builders, and network designers who possess fundamental knowledge of networking protocols, distributed systems, and basic AI/ML workflows, alongside cloud solution architects and advanced computer science students.
Shankar Ramanathan is currently a senior director of engineering at Oracle Cloud Infrastructure, where he leads the design and delivery of cloud networking services that power Oracle’s public cloud. His expertise spans end-to-end platform delivery (“chip-to-ship”), including data center-class operating systems, packet forwarding engines, virtualization technologies, and software-defined network stacks, engineered for 99.99% infrastructure availability. Over a career that includes engineering roles at Google, Cisco, and Juniper, Shankar has built and operated cloud network services, delivered merchant and in-house silicon–based routing and switching systems, defined ASIC data paths for multi-chassis switching systems and distributed forwarding, and has built highly available networking services. A graduate of BITS Pilani (1998), Shankar holds four patents in QoS, multi-chassis solutions, Fibre Channel, and link aggregation technologies.
