Aviz Networks outlines AI fabric workflows with Spectrum-X and SONiC

Aviz Networks’ AI Infrastructure Field Day 2 session outlines how to configure and run AI network fabrics using NVIDIA Spectrum-X and SONiC, with Aviz ONES supporting orchestration and Day 2 operations. The coverage focuses on workflow design, QoS for lossless transport, and telemetry-driven troubleshooting across network, compute, and GPUs.

Research Overview

The session frames AI networking as requiring separate front-end and back-end traffic paths, differentiated QoS, and coordinated operation between compute and the network fabric.

It also highlights the need for closed-loop telemetry to support observing and operating AI workloads at scale.

Key Findings

The session describes enterprise gaps around automation and orchestration for accelerating deployment, service activation, and expansion across heterogeneous environments.

It presents two example approaches: adopting NVIDIA Spectrum-X through a reference architecture and building an open networking fabric using SONiC for flexibility.

Technical Breakdown

The workflow is organized across design and Day 0 to Day 2 stages, including POD topology choices, oversubscription targets, QoS domains, and failure domain planning.

For Day 0, it covers fabric bring-up, BGP control-plane activation, lossless QoS using PFC/ECN, and policy baselines.

Operational Impact

For Day 1, the session describes tenant onboarding with segmentation and RBAC, along with change-safe rollout practices.

For Day 2, it focuses on scaling operations, alert management, intent drift detection, and guided troubleshooting, including observability that correlates network, compute, and GPU health.

Product Update

The session’s demos include SONiC deployed within ONES, covering Day 0 orchestration of BGP, EVPN/VXLAN, and lossless QoS, plus Day 2 observability across the network, compute, and GPUs.

It also describes Spectrum-X with NVIDIA Air and ONES aligned to a reference-architecture bring-up, including service unit creation spanning compute and network and Day 2 operations using telemetry and SLO tracking.

Blog Signals brief is a fact-based summary of Aviz Networks’ Field Day 2 session, which details AI fabric design and Day 0–2 operational workflows using NVIDIA Spectrum-X and SONiC with Aviz ONES for orchestration and telemetry-led operations. For enterprise IT and security decision-makers, the focus is on how to structure network QoS and closed-loop observability tied to GPU and NIC health to support day-to-day operations.