AI Data Center Network Design Handbook ler

Isbn 13: 9798174480841

pdf AI Data Center Network Design Handbook

por

Escolha um formato:

Escolha um formato:

zip ler
rar ler
pdf ler
epub ler
txt ler
djvu ler

Descrição do livro

Modern AI clusters are only as fast as the network connecting their GPUs.AI Data Center Network Design Handbook gives network engineers, infrastructure architects, SREs, and technical leaders a practical framework for understanding and designing the high-performance fabrics behind modern AI training and inference.
Instead of treating the network as simple plumbing, this handbook shows you how workload behavior, GPU topology, transport protocols, optics, storage, power, cooling, and operations combine to determine whether thousands of accelerators operate as one efficient system—or spend expensive time waiting on each other.
Inside, you’ll learn how to:

  • Design spine-leaf and Clos fabrics for high-bandwidth AI workloads
  • Understand rail-optimized, dragonfly, torus, and hierarchical topologies
  • Compare RoCEv2 and InfiniBand for RDMA-based GPU communication
  • Plan bisection bandwidth, oversubscription, switch radix, and fabric capacity
  • Manage PFC, ECN, DCQCN, incast, microbursts, and lossless Ethernet
  • Understand NCCL, all-reduce, all-gather, and collective communication behavior
  • Evaluate copper, multimode fiber, single-mode fiber, transceivers, and co-packaged optics
  • Architect storage paths using parallel file systems, NVMe-oF, and GPUDirect Storage
  • Build telemetry, fault-isolation, and self-healing network operations
  • Co-design networking with power delivery, liquid cooling, security, and facility constraints
The book contains 26 chapters across eight parts, worked sizing examples, topology discussions, operational guidance, and clear explanations designed to remain useful even as specific hardware generations change.Build a stronger mental model of AI infrastructure—and design networks that keep expensive GPUs productively working.

Número de páginas :121
Isbn 13 :9798174480841
Encadernação AI Data Center Network Design Handbook:Capa Comum