HN836: Optimize AI Traffic on the WAN with Ciena’s Integrated IP Networking (Sponsored)

The episode begins by establishing that AI inferencing—the deployment and consumption phase of AI models by enterprises and end users—represents a fundamentally different traffic pattern than traditional broadband or training workloads. Rafael Francis and Vinny Santos from Ciena explain that while AI inferencing increases bandwidth demands, especially with multimodal (video, image) content, the core challenge is not simply adding capacity everywhere. Instead, service providers must optimize for multiple competing constraints: latency (often more critical than compute latency), deterministic performance, always-on availability for agentic AI, and multi-cloud connectivity where traffic flows across diverse providers rather than following traditional north-south data center patterns.

The hosts discuss how AI inferencing traffic represents an amplification of existing network challenges rather than entirely new problems. Multi-cloud deployments, for instance, require enterprises to interconnect multiple specialized providers, creating significant east-west traffic flows that demand sophisticated traffic engineering. The solution cannot be solved through raw bandwidth increases alone; instead, it requires intelligent path selection and dynamic adaptation.

At the IP layer, Ciena advocates for Segment Routing with FlexAlgo as a modern alternative to legacy MPLS technologies like RSVP and LDP. FlexAlgo creates multiple virtual topologies within a single physical network, allowing segment routing to dynamically select paths based on real-time constraints like latency. The example given involves a financial trading infrastructure customer where a service provider network failure causes unacceptable latency increases; through FlexAlgo and latency measurement via TWAMP or Y1731, traffic can automatically reroute to the lowest-latency path without human intervention.

A key distinction emerges between distributed path computation (via FlexAlgo within routing domains) and centralized computation (via Path Computation Elements for multi-domain scenarios). The speakers emphasize that AI agents operating 24/7 require networks that adapt at machine speed—far faster than traditional human-driven configuration management. Legacy technologies like RSVP are described as too rigid and complex for this autonomous, near-real-time responsiveness.

FlexEthernet (FlexE) is presented as a complementary underlay technology that provides service isolation and latency reduction by functioning as a Layer 1.5 technology, enabling TDM-like guarantees within converged IP infrastructure. The integration of these capabilities—segment routing, FlexAlgo, and FlexE—allows service providers to combine the flexibility of modern IP with the performance guarantees historically associated with optical networks.

On the optical side, Ciena emphasizes the integration of coherent optical modems directly into routers via coherent pluggables rather than deploying separate transponders. This eliminates operational blind spots between IP and optical layers while reducing power consumption and latency. The discussion of Shared Risk Link Groups (SRLGs) highlights a critical operational benefit: understanding that two supposedly diverse IP paths might share the same physical fiber bundle, a visibility that only comes from integrated IP-optical awareness.

Telemetry is positioned as foundational to this closed-loop automation. Streaming telemetry via gRPC and GNMI, combined with performance metrics from TWAMP and Y1731, feeds data into control systems that drive autonomous network adjustments. The speakers stress that telemetry without control is merely a dashboard, while control without telemetry is flying blind.

Ciena's software offerings—Navigator (multi-layer controller for Ciena domains) and Blue Planet (multi-vendor, multi-domain orchestration)—enable this automation. Navigator understands relationships between IP and optical layers simultaneously, eliminating operational silos. Blue Planet extends to OSS/BSS layers and spans multiple vendor domains. Both integrate AI operations capabilities for troubleshooting and predictive analytics.

The role of AI in network operations is carefully positioned: it currently excels at reactive troubleshooting and root cause analysis across multi-layer environments but does not drive routing decisions. Routing decisions remain the domain of protocols like segment routing and FlexAlgo, with AI informing these decisions through trend analysis and failure prediction. The speakers reject the notion of fully autonomous networks running without human oversight, instead describing a spectrum of autonomy where operators retain decision authority while gaining faster insights and more precise tools.

Security is addressed through multiple layers. MACsec provides link-layer or end-to-end encryption on high-speed WAN links (400Gbps, 800Gbps), while WaveLogic coherent modems support bulk encryption at optical layer without latency penalty. Post-quantum cryptography considerations are emerging, with manually pre-shared keys and AES-256 offering quantum-safe properties.

The Ciena portfolio is positioned as comprehensive: capacity via coherent routers (up to 1.16 Tbps bandwidth mode) with integrated pluggables, intelligence via SAOS operating system (including segment routing, EVPN, FlexAlgo, FlexE, MACsec), and orchestration via Navigator and Blue Planet. The company sells to service providers, hyperscalers, enterprises, and vertical markets (financial, healthcare, utilities, research). The speakers emphasize avoiding the misconception that Ciena is only an optical vendor; the routing and IP layer capabilities are equally significant for AI-ready networks.