Cisco and Aviz Networks Discuss Efficient AI Networking Strategies

Recent insights from a joint bootcamp by Cisco and Aviz Networks highlighted network design strategies essential for managing Artificial Intelligence (AI) workloads. Key discussions encompassed the transition to AI-centric architectures and the necessary technological advancements to sustain high performance and low latency.

Cisco 8000 Series and SONiC

During the event, Cisco product management discussed how the Cisco 8000 series leverages Silicon One technology along with SONiC to support the requirements of modern AI infrastructures. The presentation emphasized the use of Remote Direct Memory Access (DMA) (RDMA) over Converged Ethernet (RoCE v2), Power Factor Correction (PFC), and Explicit Congestion Notification (ECN) for managing heavy Graphics Processing Unit (GPU) traffic efficiently.

Observability with Aviz ONES

Aviz’s Open Networking Enterprise Suite (ONES) was demonstrated as a solution for monitoring network performances, examining congestion issues, and ensuring adherence to Quality of Service (QoS) policies in multi-vendor settings. Attendees also reviewed approaches to constructing scalable leaf-spine network topologies vital for AI activities.

Conclusion

The bootcamp emphasized the importance of implementing flexible network architectures that support AI workloads, focusing on lossless connectivity and effective congestion management. This overview represents a succinct summary of the original blog post, which detailed valuable strategies for decision-makers in the field.