Cisco Live US 2025 (San Diego)#
Enrichment note: the Cisco Live on-demand library is a login-gated JavaScript app; page fetches returned only the site shell, so every write-up below is grounded solely in the program abstracts. Vendor-program caveat applies throughout — Cisco speakers presenting Cisco products — and each entry flags where the substance looks general versus product-specific.
June 8, 2025 – June 12, 2025 Program
Core#
400G, 800G, and Terabit Pluggable Optics: What You Need to Know - BRKOPT-2699#
Mark Nowell (Cisco), Errol Roberts (Cisco)
- Why: The optics roadmap is now the binding constraint on DC fabric design — power, reach, and breakout options determine what topologies are even buildable. Nowell is a long-standing IEEE 802.3 optics contributor, which makes this a standards-insider survey rather than a datasheet read.
- What: A survey of current 400G pluggable optics and their applications, next-generation 800G/1.6T form factors including breakout use cases, the status of 800 Gb/s and 1.6 Tb/s Ethernet standardization, and co-packaged optics (integrating optical interfaces into switch ASICs), framed by industry dynamics and AI-driven demand (from abstract only).
- How: Technology survey across form factors, IEEE standardization tracks, and co-packaging directions, covering both coherent and noncoherent optics.
- Where applicable: Broadly applicable to anyone specifying optical interconnect for DC or DCI — the standards and form-factor content is vendor-neutral even though the speaker is Cisco. Assumes working deployment knowledge of optical modules.
Advancements and Applications of Coherent Pluggable Technology in High Bandwidth Transport Networks - BRKOPT-2556#
Anuj Malik (Cisco)
- Why: Coherent pluggables are collapsing the boundary between the IP and optical layers — directly relevant to DC/WAN interconnect, where the same pluggable now serves router ports and line systems.
- What: The technical underpinnings of coherent pluggables — tunable lasers, high-speed ADC/DACs, DSP algorithms — plus performance comparisons (data rates, reach, power consumption) against traditional coherent solutions, real-world deployment scenarios, and the trend toward IP/optical convergence (from abstract only).
- How: Coherent DSP packaged into pluggable transceiver form factors, compared against embedded coherent line systems on rate/reach/power metrics.
- Where applicable: High-bandwidth transport and DCI. Vendor talk, so expect the case studies to favor Cisco outcomes, but the component physics and metric trade-offs are general.
Ethernet Evolution and Challenges in the world of AI - BRKNWT-2504#
Kapil Mehta (Cisco)
- Why: This is the lossless-fabric problem the PM works on daily — RDMA and elephant flows breaking Ethernet's assumptions in AI back-end networks — plus a read on where Ultra Ethernet Consortium standardization is actually heading.
- What: Ethernet's evolution history, its current failure modes under AI switching workloads (RDMA, elephant packet flows), Cisco's solutions, and the innovations being standardized by the Ultra Ethernet Consortium (from abstract only).
- How: Survey of the mechanisms being applied to congestion and losslessness at AI scale, positioned against UEC's standardization work items.
- Where applicable: AI back-end networks generally; the problem framing is industry-wide, but the "solutions" portion will be Cisco-flavored.
How Cisco build AI readiness IP Backbone with 8000 Series - BRKMSI-3000#
Rob Piasecki (Cisco), Ruben Lopes da Fonte (Cisco)
- Why: Rare config-level treatment of lossless fabric mechanics — RoCEv2, PFC, ECN — including CLI verification, which is exactly the lossless-fabric internals the PM wants, not just architecture slideware.
- What: How Cisco builds AI infrastructure for webscalers: the requirements, the RoCEv2/PFC/ECN technology, detailed IOS-XR configuration, CLI verification that PFC and ECN are working, and design guidance (from abstract only — note the abstract itself is typo-laden, suggesting a hands-on rather than polished session).
- How: Step-by-step IOS-XR configuration of PFC and ECN with CLI validation, wrapped in design recommendations for AI-ready IP backbones.
- Where applicable: Platform-specific to Cisco 8000-series/IOS-XR, but the design principles (ECN threshold tuning, PFC behavior verification, lossless design patterns) transfer to any RoCEv2 fabric.
Optimizing AI-Ready WAN Infrastructure with Microsoft and Cisco - CSSENT-2100#
Sarav Subramanian (Microsoft), Ruben Lopes da Fonte (Cisco), Gaisin Rustam (Microsoft)
- Why: An end-user case study from Microsoft on deploying RDMA in the Azure WAN — extending lossless-network practice beyond the DC into the WAN, which few operators talk about publicly.
- What: How AI-ready infrastructure is being built in the Azure WAN using RDMA for high throughput, low latency, and lossless transfer with minimal CPU load, and the congestion-management strategies (Priority Flow Control, ECN) used to handle oversubscription and microbursts (from abstract only).
- How: RDMA across WAN infrastructure with PFC/ECN congestion control for microburst and oversubscription scenarios.
- Where applicable: Microsoft Azure WAN-specific setup. Joint vendor-customer session, so expect some marketing gloss, but the scenario — lossless behavior at WAN scale — is unusually candid territory for Microsoft to discuss.
Groq: An AI Success Story - Deploying High-Performance, and Low-Latency AI Networks Using Cisco Nexus 9000 - CSSDCN-2005#
Swetha Velamuri (Cisco), Cameron Ferdinands (Groq)
- Why: A real customer (Groq) describing inference-network design — inference traffic has different characteristics from training (latency-critical, bursty, iterative) and public detail on inference fabrics is scarce.
- What: How Groq uses Nexus 9000/NX-OS to achieve scalability, bandwidth, low latency, and congestion management for AI inference, plus how Cisco optics choices reduce cost and energy (from abstract only — the abstract makes performance and cost claims without numbers; treat specifics as unverified until seen in the recording).
- How: Groq's software-first hardware approach paired with Nexus congestion management and optics innovations.
- Where applicable: Groq-specific inference deployments; success-story framing with a Cisco co-speaker, so expect a product-friendly narrative rather than a post-mortem.
Cisco Nexus 9000 Architecture - BRKDCN-3939#
Faraz Taifehesmatian (Cisco)
- Why: Switching silicon and forwarding internals are squarely in the PM's core interest; this is the deep-dive on the ASIC generation that underpins the AI-fabric sessions above.
- What: An in-depth architecture study of the latest Nexus 9000 modular and top-of-rack switches: ASIC architecture, forwarding mechanics, physical design elements, and the hardware-enabled features that deliver DC network services (from abstract only).
- How: Architecture walkthrough from silicon to system.
- Where applicable: Cisco-specific silicon, but the trade-offs discussed (buffering, forwarding pipeline, physical design) are transferable to evaluating any modern DC switch ASIC.
VXLAN EVPN: A Deep Dive into Packet Forwarding - BRKDCN-3966#
Lukas Krattiger (Cisco)
- Why: "A day in the life of a packet" is exactly the datapath-internals treatment the PM values — most EVPN content covers control plane; this follows the data plane.
- What: How routing and bridging actually happen across a VXLAN EVPN fabric — what occurs from the ingress VTEP through to the egress VTEP and beyond, including how multi-destination traffic is handled (from abstract only).
- How: Step-by-step packet walk through the overlay, ingress to egress.
- Where applicable: The mechanics are general to VXLAN EVPN regardless of vendor; examples will be NX-OS-based.
SONiC Architecture and Deployment Deep Dive - BRKMSI-2004#
Jiri Chaloupka (Cisco)
- Why: SONiC is where open-source NOS momentum actually is, and a Cisco deep dive covering SONiC on their own Silicon One dataplane is a strong signal about where the disaggregation story is heading — including AI back-end networks as a deployment driver.
- What: Part one covers SONiC's OS architecture in detail with examples of SONiC running on Cisco Silicon One NPU/data plane; part two covers deployment examples and when SONiC should be considered, including AI frontend and back-end networks (from abstract only).
- How: Architecture decomposition of SONiC's componentized design, followed by deployment-decision guidance.
- Where applicable: SONiC on Cisco hardware specifically, but the architecture content and the "when to choose SONiC" framing apply to any disaggregated-NOS evaluation.
Deployment of Micro-Segmentation in Cisco NX-OS VXLAN EVPN Fabrics with VXLAN Group Policy Option (GPO) - BRKDCN-2633#
Max Ardica (Cisco)
- Why: East-west segmentation enforced in the fabric itself — the intersection of the PM's network security and DC overlay interests — and grounded in an IETF draft rather than a purely proprietary mechanism.
- What: Micro-segmentation using security groups and security group ACLs in NX-OS VXLAN EVPN fabrics via VXLAN Group Policy Option (IETF draft-smith-vxlan-group-policy-05), covering the control- and data-plane components of the implementation and common use cases including service-chaining simplification (from abstract only).
- How: Group policy conveyed in the VXLAN header so per-application policy follows the workload anywhere in the fabric, replacing perimeter-centric inspection.
- Where applicable: NX-OS EVPN fabrics; the mechanism follows an IETF draft so the concept is portable, but the described implementation is Cisco-specific.
Security superpowers with eBPF and Tetragon - BRKSEC-2167#
Liz Rice (Cisco)
- Why: Liz Rice is one of the leading voices in eBPF/Cilium, and Tetragon is the open-source enforcement engine underneath Cisco's Hypershield — this session reveals where a major vendor's security stack is actually being built, from a contributor rather than a marketer.
- What: A deep dive into Tetragon: YAML tracing policies mapped to eBPF programs attached to a variety of event types, multi-program "following" of a series of events to pinpoint likely-malicious behavior, and in-kernel event aggregation/filtering to minimize observability overhead (from abstract only).
- How: eBPF programs attached at kernel event points; policy-based correlation across event sequences; in-kernel aggregation and filtering to keep overhead low.
- Where applicable: Linux hosts and Kubernetes broadly — Tetragon is open source and vendor-neutral in applicability, while also being the substrate for Cisco's commercial Hypershield.
The Swiss Army Knife of Cloud Native Networking - BRKCLD-1012#
Raymond De Jong (Cisco)
- Why: Hits the PM's Cilium/BGP/IPAM interests directly — BGP from Cilium and LB IPAM for multi-tenant pools are the operational pieces most teams hit in production.
- What: Cilium's advanced features — BGP, load balancing, LB IPAM, and network policies — for multi-cloud and multi-tenancy networking, including routing/connectivity between Kubernetes clusters and external networks, BGP setup across different topologies, and multi-tenant routing with multiple IP pools for LoadBalancer IP assignment (from abstract only).
- How: eBPF datapath (Cilium) with a BGP control plane for external reachability, and LB IPAM pools giving platform engineers per-tenant LB IP allocation.
- Where applicable: General Cilium/Kubernetes environments; mechanisms are open source even though the speaker is Cisco.
Are You Prepared for the Next Typhoon? - BRKSEC-2499#
Paul Giralt (Cisco), Steve Nowell (Cisco)
- Why: Salt Typhoon is the defining nation-state compromise of network infrastructure in this era, and this session combines Talos threat insight with concrete hardening guidance plus Cisco's "secure by default" roadmap — with an explicit feedback opportunity on that roadmap.
- What: The nature of the 2024 Salt Typhoon intrusions across multi-vendor infrastructure at major telecoms as understood by Talos, specific hardening steps available today, and planned future "secure by default" enhancements to network infrastructure (from abstract only).
- How: Attack-surface analysis followed by hardening best practices and defensive-design changes.
- Where applicable: Broadly applicable to network infrastructure security; telecom operators are the most-affected vertical, but the hardening lessons apply to any operator of carrier or enterprise network gear.
Innovative Authentication: Beyond Passwords - BRKSEC-2584#
Stefan Duernberger (Cisco)
- Why: Directly the identity stack the PM tracks — passkeys, OAuth 2.0/OIDC — plus two genuinely emerging items (Google's Device-Bound Session Credentials and Shared Signals and Events) that sit beyond the basics.
- What: The role and implementation of passkeys, OAuth 2.0 and OIDC for cross-platform authentication, Google DBSC, Cisco Duo Passport, and an overview of Shared Signals and Events — all framed within a Zero Trust architecture (from abstract only).
- How: Protocol survey connecting credential technology (passkeys/WebAuthn) to authorization frameworks (OAuth/OIDC) to session-integrity (DBSC) and continuous-signal (SSSE) mechanisms.
- Where applicable: General web and enterprise authentication; the Duo Passport portion is Cisco product content, but the protocol material is standards-based and general.
Deep Dive into Cisco's Use of QUIC, MASQUE and OS Native Capabilities to Deliver Frictionless Zero Trust Access - BRKSEC-3027#
Vinny Parla (Cisco)
- Why: Shows where identity and transport converge: MASQUE/QUIC-based access proxied through OS-native clients from Apple, Google, and Samsung, with WebAuthn and passkeys in the loop — a credible picture of how frictionless ZTNA gets to Wi-Fi-level ubiquity.
- What: How industry-standard protocols (MASQUE, QUIC, WebAuthn, passkeys, private 5G) can deliver Zero Trust Access with minimal user/admin friction; how native OS clients from Apple and Samsung and the Cisco Zero Trust Access module in Cisco Secure Client leverage MASQUE and QUIC for a seamless experience (from abstract only).
- How: MASQUE-style proxying over QUIC tunnels integrated with OS-native network clients and passkey-based user identity.
- Where applicable: Enterprise and consumer device access; heavily ecosystem-dependent (collaboration with Apple, Google, Samsung), so the OS-client pieces are platform-specific while the protocol layer is standard.
Adjacent#
Impact of AI Traffic in Transport Networks - BRKSPG-1180#
Virginia Teixeira (Cisco)
- Why: Near-core for anyone responsible for DC/WAN interconnect: AI traffic patterns (RAG, grounding, agents, edge distribution) change DCI and WAN dimensioning assumptions in ways transport teams haven't absorbed yet.
- What: An analysis of AI traffic characteristics across a set of AI use cases and their impact on WAN and DCI design requirements (from abstract only).
- How: Traffic-pattern characterization mapped onto transport network design levers.
- Where applicable: Service-provider and large-enterprise transport. The speaker's own framing is honest that AI deployment models are still settling, so expect early directional analysis rather than settled engineering rules.
Multi-Tier Fabrics: Network Designs for the Modern Data Center - BRKDCN-2999#
Marina Ferreira (Cisco)
- Why: The back half of this session is genuinely relevant physical-DC content — super-spine and hyperscale plane designs, where 400G/800G is cost-effective, fixed vs. modular trade-offs, and routed fabrics for cloud-native applications. The front half is Clos 101, which the PM doesn't need; hence adjacent rather than core.
- What: Fat-tree, Clos, and leaf/spine foundations extended to super-spine and hyperscale plane designs, cost-effective use of 400G/800G, fixed-vs-modular switch decisions, and a review of self-organized routed fabrics for cloud-native applications (from abstract only).
- How: Design-pattern survey with cost/scale reasoning.
- Where applicable: General data center design, vendor-neutral in principle; the routed-fabric segment connects to the same IP-underlay-for-cloud trend the PM sees in Kubernetes environments.
Connecting Your Data to LLMs: Identity, Functions and Agents are all you Need - BRKETI-2411#
Guillaume De Saint Marc (Cisco)
- Why: Agentic identity is one of the PM's explicit interests and almost absent from vendor programs — this is the closest hit: per-request authorization deciding which data an agent may fetch and serve.
- What: Architectures and tools for connecting enterprise data (including on-prem) to LLMs without data migration, including real-time prompt inspection to identify the data required by a request, determine whether the user has rightful access, and serve only authorized data (from abstract only).
- How: Prompt-time authorization — mapping an incoming prompt to data requirements and checking entitlements before data is served to the model.
- Where applicable: Enterprise GenAI deployments; Cisco brings product positioning, but the underlying pattern — entitlement-aware data serving for agent requests — is broadly relevant to anyone designing agentic identity.
EVPN Fabric in the Real World: Securing Critical Infrastructure in Finance and Aviation - BRKENS-2655#
Daniel Cantor (Fiserv), Ekta Ghosh (Adani Airports Holding Limited), Rahul Kachalia (Cisco)
- Why: Two real end users (Fiserv, Adani Airports) discussing EVPN-fabric modernization under regulatory pressure. Worth a look for the case-study value, but the abstract is executive-framed ("high-impact executive session") and offers no metrics, failure discussion, or technical depth — approach with skepticism.
- What: How Fiserv and Adani Airports used Catalyst networking and EVPN fabric as the foundation for automated, always-on infrastructure, with claimed reductions in downtime and improved security posture (from abstract only — no numbers in the abstract; verify specifics in the recording).
- How: Executive discussion/Q&A format per the abstract.
- Where applicable: Regulated enterprise environments (financial services, aviation/airports); campus and DC-edge scale rather than hyperscale.
Wildcard#
Packet Buddy - Chatting with your PCAPs, via the Cloud or Locally on Your Own Machine! - AIHUB-2000#
Dave Zacks (Cisco), Prapanch Ramamoorthy (Cisco)
- Why: LLM-assisted packet analysis is a genuinely new tool category, and the PM lives in PCAPs — this is the kind of thing worth knowing exists before it lands in a commercial product.
- What: An open-source solution ("PACKET BUDDY") that lets you chat with your own Wireshark captures via a choice of LLMs — asking what traffic is present, what errors it sees, and getting help understanding packets and flows — deployable via cloud or locally on your own machine (from abstract only).
- How: PCAP ingestion fed to an LLM with conversational querying; the local deployment option is the privacy-relevant detail.
- Where applicable: Any environment; open source, with the obvious caveat that feeding captures to cloud LLMs has data-exposure implications — the local mode matters.
Packets in Space: Networking for the Final Frontier - CISCOU-1057#
Lexie Cooper (Blue Origin), Eric Peterson (Blue Origin), Patrick Garggan (Cisco)
- Why: Designing networks where failure is not an option and latency requirements are extreme forces first-principles thinking that is useful calibration for anyone building "critical" terrestrial infrastructure.
- What: A moderated Q&A with Blue Origin "rocket network engineers" on designing mission-critical networks for rockets and spacecraft — real-time telemetry, low-latency requirements, and extreme environmental conditions (from abstract only — note the program abstract is truncated mid-sentence, so the full scope of the session is unclear from the text alone).
- How: Q&A format rather than a prepared technical talk.
- Where applicable: Aerospace-specific, but the deterministic, failure-intolerant design lessons generalize to any network with hard availability and timing requirements.
An Introduction to Quantum Mechanics, Computing, and Networking - BRKETI-1401#
Tim Szigeti (Cisco)
- Why: Cisco is making a serious R&D bet on quantum networking, and this is the clearest signal in the program of where the long-horizon network and security landscape may move (distributed quantum computing, entanglement-based secure communications, clock sync).
- What: An introduction to quantum mechanics focused on qubits and their special properties — superposition, entanglement, no-cloning, including mathematical proofs — an overview of quantum computing's impact on network communications and security, and the building blocks, opportunities, and challenges of quantum networks, plus Cisco's current work (from abstract only).
- How: Tutorial format; foundational science first, then networking implications.
- Where applicable: Nothing here is deployable today — it is forward-looking R&D context, useful for spotting the trajectory rather than near-term engineering.
Themes#
- AI fabrics are the gravitational center of the entire program. Optics, Ethernet standards, switch ASICs, IOS-XR configuration, WAN design, and even coherent transport are all being reframed around AI training/inference traffic. The lossless-Ethernet problem (RoCEv2, PFC/ECN, microbursts, elephant flows) and the Ultra Ethernet Consortium's standardization work are the technical through-line — and notably, the lossless conversation has moved from the DC back-end into the WAN (Microsoft's Azure RDMA session).
- Optics is now a first-class systems constraint. 800G/1.6T pluggables, breakout economics, 800 Gb/s and 1.6 Tb/s Ethernet standardization, and early co-packaged optics show the industry treating the optical module — its power, reach, and form factor — as a determinant of fabric topology rather than a pluggable afterthought.
- eBPF has crossed from OSS novelty to vendor substrate. Liz Rice presenting Tetragon at Cisco Live, Tetragon powering Hypershield, and a full session on Cilium BGP/LB IPAM indicate that eBPF-based networking and security is now the foundation of a major vendor's cloud-native and security strategy — a meaningful trajectory signal for the ecosystem.
- Identity is converging on session-bound credentials and continuous signals. Passkeys/WebAuthn, OAuth/OIDC, Google's Device-Bound Session Credentials, Shared Signals and Events, and MASQUE/QUIC-based zero trust collectively point away from one-time login events toward continuous, protocol-level session integrity — with agentic/LLM data-access authorization just beginning to appear at the edge of the program.
- Nation-state infrastructure threat is driving a "secure by default" agenda. Salt Typhoon gets a full session combining Talos threat detail, concrete hardening steps, and a vendor roadmap for secure-by-default network devices — network device security is being treated as a lifecycle discipline, not a hardening checklist.
- Open NOS momentum keeps building. A SONiC architecture deep dive on Cisco's own Silicon One dataplane — with AI back-end networks named as a key deployment driver — shows even incumbent vendors treating disaggregated operating systems as a primary path, not an experiment.
About this digest
ToFaRev picked and wrote this digest on Nebius Token Factory, from the conference's published program. A person reviewed it in a pull request before it was published. Write-ups marked "from abstract only" had no fuller source to read. Source: digests/ciscolive-us-2025.md · prompt: prompts/ciscolive-us-2025.md