ITNOG10#
Source depth note: ITNOG10 published no abstracts — the program lists only titles, speakers, and slide links, and slide content was not retrievable in this run. Every write-up below is therefore grounded in the session title (and slide-link filename) alone; items are marked (title only). Where a mechanism is described with "presumably" or "expect," it is an inference about what the talk likely covers, not a claim about its contents — verify against the linked decks before acting on it.
April 20, 2026 – April 21, 2026 Program
Core#
Integration of Security Functions in EVPN/VXLAN Using Group-Based Policy#
Nicola Modena
- Why: Sits exactly at this PM's intersection of DC fabric and network security: enforcing policy inside the EVPN/VXLAN overlay rather than bolting on perimeter appliances.
- What: An approach to integrating security functions into EVPN/VXLAN fabrics using Group-Based Policy. (title only)
- How: GBP as a technology tags traffic with policy-group identifiers carried in the VXLAN header so the fabric itself enforces segmentation; expect the talk to cover where security functions hook into that model — the deck has the specifics.
- Where applicable: EVPN/VXLAN DC fabrics. GBP support is uneven across NOS/silicon (strongest on Cisco platforms), so check which parts are portable before assuming multi-vendor applicability.
National Layer-2 wholesale FWA leveraging VxLAN#
Christian Biasibetti
- Why: A rare case study of applying DC overlay tech (VXLAN) to a national wholesale FWA backhaul — a transferable pattern for L2 service delivery over an IP core, which is squarely in the PM's fabric/interconnect space.
- What: Design and operation of a national layer-2 wholesale service for Fixed Wireless Access, built on VXLAN. (title only)
- How: Presumably VXLAN tunnels extending L2 domains from aggregation points toward FWA endpoints across an IP core; the deck carries the actual architecture.
- Where applicable: Telco wholesale/FWA context, so scale numbers may be footprint-specific; the L2-service-over-VXLAN pattern itself is broadly reusable.
BGP nell'underlay multi-cloud: resilienza e simmetria tra on-prem e public cloud#
Carlo Pio Luigi Ciciriello
- Why: Directly the PM's cloud-native networking and DC/WAN-interconnect territory: BGP design for hybrid underlays, where resilience and return-path symmetry are the genuinely hard parts.
- What: A BGP underlay design spanning on-prem DCs and public clouds, targeting failover resilience and traffic symmetry between the two. (title only)
- How: Expect session/policy design and path selection across dedicated cloud interconnects to keep flows symmetric on both failover and normal operation — the classic pitfall this class of talk addresses.
- Where applicable: Broadly applicable to any hybrid/multi-cloud design using dedicated interconnects (Direct Connect/ExpressRoute-style), operator or enterprise.
Open Networking and IPoDWDM in Production: Operational Lessons from European Networks#
Alessandro Giudice
- Why: IPoDWDM with 400G ZR+ pluggables (per the slides filename) is the current frontier of router-integrated coherent optics for DCI/WAN interconnect — and production lessons from real European networks, not vendor spec sheets, are the valuable part.
- What: Operational experience running open networking / IPoDWDM (400G ZR+) in production across multiple European networks. (title only)
- How: Coherent 400G ZR/ZR+ pluggables terminated in router ports, collapsing the transponder layer; expect real-world issues around optical budgets, multivendor interoperability, and troubleshooting.
- Where applicable: DCI and WAN interconnect generally. The multi-network framing suggests lessons transfer beyond a single operator's setup.
Hollow Core Fibre - Nothing is better than glass for guiding light#
Steve Jones
- Why: Hollow-core fiber is the most-watched optics disruption (ultra-low latency, low nonlinearity), and the title reads as a deliberately contrarian, pro-solid-glass take — exactly the skeptical operator perspective worth hearing before buying into the hype. Pure PM optics territory.
- What: An assessment of hollow-core fiber versus conventional solid glass fiber for guiding light. (title only)
- How: Presumably a physics-and-operations comparison — latency, attenuation, nonlinearity, cost, and deployment maturity; verify the framing in the deck.
- Where applicable: Long-haul, subsea, and latency-sensitive DCI routes; the practical constraints are reach, cost, and maturity of supply.
SRv6 uSID: Deterministic AI Backend Networking and Unified DC Frontend Design#
Alberto Donzelli — the slides filename ("Cisco-final") indicates a Cisco-affiliated session; treat as a vendor-informed design pattern, not a neutral case study
- Why: AI backend networks are the hottest fabric-design problem right now (deterministic, lossless behavior for collective/parallel traffic) — squarely in the PM's RDMA/lossless-fabric interest.
- What: A design combining SRv6 micro-segment (uSID) networking for AI backend traffic with a unified frontend/DC architecture. (title only)
- How: uSID compresses many segments into a single IPv6 address for fine-grained steering; the claim to verify in the deck is how SRv6 delivers "deterministic" backend behavior and how frontend and backend unify under one addressing scheme.
- Where applicable: Requires SRv6/uSID-capable silicon and NOS, so the mechanism is partially vendor/silicon-bound; the AI-backend problem framing transfers across the industry.
MIX DC3#
Mauro Magrassi — MIX (Milan Internet exchange)
- Why: A buildout talk for an IXP's third datacenter — pure physical-DC territory (space, power, topology, fabric, inter-site connectivity) at a facility whose uptime matters to the Italian internet.
- What: Presentation of MIX's new DC3 facility. (title only — no abstract; likely design and service roadmap for the new site)
- How: Unknown from the program; expect site selection, fabric/topology design, and how DC3 interconnects with the existing MIX estate.
- Where applicable: IXP/colo context; DC buildout lessons (power density, topology, inter-DC links) are broadly relevant.
Inside Seabone: experiences of a Tier-1 Backbone#
Gabriele Laghi, Patrizia Rossi, Antonio Ferretti
- Why: A Tier-1 operational tell-all is exactly the "failure stories and hard-won lessons" category this PM up-ranks — Tier-1 backbones see failure modes no one else does.
- What: Behind-the-scenes operational experiences running the Seabone Tier-1 IP backbone. (title only)
- How: Unknown; expect scale, peering, capacity engineering, and incident practice from the deck.
- Where applicable: Tier-1/transit scale — some lessons are inherently scale-dependent, but the operational patterns (maintenance, convergence, tooling) transfer.
Migrazione del nostro Backbone a SR-MPLS perdendo pochi ping...#
Denis Corradini
- Why: A migration story with the packet-loss count right in the title — the most transferable kind of operator talk: how to move a live backbone to SR-MPLS while (nearly) not dropping traffic.
- What: Migration of an operator backbone to SR-MPLS, with impact quantified by ping loss during cutover. (title only)
- How: Expect an SR-MPLS underlay with TI-LFA fast reroute and a phased cutover; the durable value is the cutover method and the measured loss, not the product choice.
- Where applicable: Any MPLS-capable backbone — a broadly applicable migration pattern, not org-specific.
Transforming BGP operations to serve the best prefix out of 26 million#
Alessandro D'Eliseo, Antonio Prado
- Why: BGP operations at 26-million-prefix scale is the extreme end of BGP tooling and datapath internals — exactly what the PM tracks in software networking.
- What: How BGP operations were reworked to select and serve the "best" prefix out of a 26-million-prefix table. (title only)
- How: Unknown; at that scale this is necessarily automation and policy engineering around massive tables — the deck has the architecture.
- Where applicable: The 26M figure signals a very large content/anycast or transit platform, so the scale is org-specific; the tooling and automation patterns partially transfer.
Splitting ottico passivo per ridundanza link#
Mirko Mosca
- Why: A physical-optics technique for link redundancy using passive splitters — cheap, unpowered redundancy directly in the PM's optics/cabling territory, and the kind of low-tech trick high-end roadmaps overlook.
- What: Using passive optical splitting to provide link redundancy. (title only)
- How: Presumably a splitter feeds two paths/receivers so a failure on one path does not drop the link; the design constraint to check is the splitter's insertion loss (roughly 3 dB per branch) eating into the loss budget.
- Where applicable: Short-reach/DCI links and cost- or power-constrained sites; loss-budget math limits reach, so it does not scale to long spans without amplification.
Adjacent#
Building the longest SR-TE lsp in the world#
Umberto M
- Why: A boundary-pushing SR-TE experiment — useful mainly for understanding SR datapath scaling limits (SID-stack depth, MTU, resilience) even if you never build one.
- What: Construction of a claimed record-length SR-TE LSP. (title only)
- How: Presumably an intercontinental path stitched across cooperating networks; the engineering constraints hit along the way are the transferable content.
- Where applicable: Depends on participating ASes; the limits discovered are general to SR-TE deployments.
OTN oltre il trasporto: reti programmabili, latenze garantite e architetture quantum-safe#
OPNET — vendor session (OPNET listed as the speaker)
- Why: Signals where optical transport is heading: programmable OTN, guaranteed-latency services, and post-quantum crypto in transport — the last item touching the PM's security interest.
- What: A vendor vision for OTN "beyond transport": programmability, latency guarantees, and quantum-safe architectures. (title only)
- How: Unknown; expect a roadmap rather than operator practice.
- Where applicable: Treat as a directional market signal with a pitch component, not an adoptable technique.
Open Lab: A collaborative multi-vendor platform for experimentation and validation of next-generation networking technologies#
Flavio Luciani
- Why: A community-run, multi-vendor validation lab is infrastructure other operators can actually use — pre-production validation is a persistent gap, and shared labs are one of the few real answers.
- What: A collaborative platform for experimenting with and validating next-generation networking technologies across vendors. (title only)
- How: Multi-vendor lab environment; the access/participation model is presumably described in the deck.
- Where applicable: Broadly adoptable by operators; community-run, so watch for Italy/EU scoping.
AS/IXP Traffic Observability / Osservabilità del traffico AS/IXP#
Luca Deri, Federico Santulli
- Why: Traffic observability at AS and IXP scope — tooling others can adopt, feeding capacity planning, peering decisions, and abuse/anomaly detection.
- What: Approaches and tools for observing traffic at AS and IXP level. (title only)
- How: Unknown; expect flow-based analysis and open tooling in the deck.
- Where applicable: Any AS or IXP operator; broadly applicable.
Semplificare l'automazione multi-vendor: Client RESTCONF auto-generati dai moduli YANG#
Matteo Colantonio
- Why: Auto-generating RESTCONF clients from YANG models is a concrete, adoptable automation technique that attacks multi-vendor fragmentation at its root — exactly the kind of tooling others can pick up.
- What: A pipeline that generates RESTCONF client code from YANG modules, simplifying multi-vendor automation. (title only)
- How: YANG modules parsed into generated, typed client APIs per device/vendor, abstracting RESTCONF differences behind generated drivers.
- Where applicable: Any environment with YANG/RESTCONF-capable devices; vendor-neutral by construction, so broadly applicable.
Wildcard#
Antares: Multi-Agent AI for Automated Network Analysis and Troubleshooting in GARR Data Centers#
Francesco Lombardo — GARR (Italian research and education network)
- Why: Agentic AI performing diagnosis on production datacenter networks is an emerging subfield — and it raises exactly the authorization and trust questions around autonomous agents acting on infrastructure that the PM's identity and security work cares about.
- What: Antares, a multi-agent AI system for automated network analysis and troubleshooting in GARR data centers. (title only)
- How: Presumably cooperating specialized agents (telemetry collection, hypothesis generation, remediation); the deck has the actual architecture and scope.
- Where applicable: R&E-network DCs today; the multi-agent ops pattern generalizes, and the guard-railing questions (what may an agent do, unattended, to a production fabric?) apply anywhere.
Measuring the public Internet#
João Damas — APNIC (per slides filename)
- Why: Not a technical advance, but the measurement infrastructure behind most public claims about Internet health, routing, and deployment — knowing how those numbers are produced (and where they're blind) is quietly valuable to anyone reasoning about the public Internet.
- What: How the public Internet is measured at global scale. (title only)
- How: Unknown; expect measurement methodology and data platforms rather than a deployable technique.
- Where applicable: General context; no environment-specific constraints.
Not selected: sponsor/community slots (HaliveOS, IXP updates, ITNOG updates, PeeringDB news), 101-level sessions (Improve IPv6 Peering), and off-domain talks (telephony KPIs/fraud analytics, telco edge computing local break, HERABIT interprovider VPN, the NAF automation map, and the "Reimagining Networking in the Age of AI" keynote, which looked like a perspective talk rather than a technical one). NAF and HERABIT were the closest near-misses if you want one more automation-framework and one more inter-AS service case study.
Themes#
- Optics is the center of gravity at this operator event: hollow-core fiber, 400G ZR+ IPoDWDM in production, passive optical splitting for redundancy, and programmable/quantum-safe OTN — even a small operator conference is dominated by the physical layer this cycle.
- Segment routing as the universal dataplane: an SR-MPLS backbone migration, a record-length SR-TE LSP, and SRv6 uSID in the datacenter — SR now spans WAN transport and AI-fabric design in the same program.
- AI appears on both sides of the network: AI backends drive deterministic fabric design (SRv6 uSID), while AI enters operations as multi-agent troubleshooting (Antares) and as a theme reshaping the engineer's role.
- The overlay abstraction has escaped the datacenter: EVPN/VXLAN with Group-Based Policy for security, and national L2 wholesale FWA over VXLAN — DC overlay tech is now the substrate for telco wholesale services too.
- Automation is consolidating around standards and shared infrastructure: YANG-generated RESTCONF clients, the NAF framework, and a community multi-vendor Open Lab — moving from per-box scripts toward generated code and collective validation.
- Hybrid multi-cloud BGP is table stakes: an operator-context talk on resilient, symmetric on-prem-to-cloud underlays signals that multi-cloud connectivity is now routine operator work, not enterprise-only.
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/itnog-10.md · prompt: prompts/itnog-10.md