Technology Deep-Dives
Engineering walkthroughs of the protocols, AI fabric architectures, and load-balancing technologies inside OcNOS (RoCEv2, DLB, EVPN-VXLAN, segment routing and more), plus the Ultra Ethernet standard. Written for network architects who want implementation detail. New to the category? Start with open networking and where IP Infusion fits.
What Is Open Networking?
Silicon, hardware and network operating system as separately sourced layers, the standards that tie them together, and where OcNOS fits.
Read deep-dive → Open networking · DefinitionWhat Is Network Disaggregation?
Separating the network operating system from open hardware built on merchant silicon, and what that changes for integration, support and validation.
Read deep-dive → Open networking · DefinitionWhat Is a Network Operating System?
Control, forwarding and management planes, and what makes a NOS open or disaggregated.
Read deep-dive → Open networking · Data centerWhat Is a White Box Switch?
Bare metal switches on merchant silicon, what they give and what they ask of you, and the OcNOS-DC qualified platforms.
Read deep-dive → Open networking · Service providerWhat Is a White Box Router?
Open routers for cell site, aggregation and core, and the OcNOS-SP qualified platforms by role.
Read deep-dive → Open networking · AI fabricAI Networking Switches
400G and 800G Ethernet switches for GPU back-end fabrics: radix, lossless transport and the qualified platforms.
Read deep-dive → Underlay · ForwardingSegment Routing: MPLS
IS-IS / OSPF as IGP, label-switched forwarding without LDP, TI-LFA fast reroute, Flex-Algorithm constraint paths.
Read deep-dive → Underlay · IPv6Segment Routing v6 (SRv6)
Segments encoded as IPv6 addresses, with no MPLS data plane. OcNOS-SP has initial SRv6 support on selected routers, with more in development.
Read deep-dive → Overlay · DC + MetroEVPN-VXLAN
BGP-signaled multi-tenant L2/L3 VPN, ESI multi-homing, symmetric IRB, mass-withdraw convergence.
Read deep-dive → Routing · Internet + DCBGP
Add-path, BGP-LU, GR, RPKI ROV, AS-set / AS-confederation.
Read deep-dive → AI Fabric · LosslessRoCEv2
PFC + ECN + DCQCN tuning for GPU collectives. Dynamic load balancing, per-priority telemetry.
Read deep-dive → Silicon · Full portfolioBroadcom Switch Silicon
The complete map of Broadcom StrataXGS (Tomahawk, Trident) and StrataDNX (Jericho, Qumran) silicon OcNOS runs on: how the families differ, the generations and their advantages, and the IP Infusion qualified open switches and routers built on each.
Read deep-dive → Silicon · 800G AI FabricTomahawk 5 Switches
BCM78900 silicon, 51.2 Tbps in one 5 nm die, and the three open 800G platforms qualified on OcNOS-DC: Edgecore AIS800-64D, UfiSpace S9321-64E / 64EO.
Read deep-dive → Silicon · 1.6T AI FabricTomahawk 6 Switches
BCM78910 and BCM78914, 102.4 Tbps on a single 3 nm die, at 64 ports of 1.6T or 128 of 800G. What a 1.6T port changes in a fabric design, and what carries over from an 800G fabric. OcNOS-DC will support Tomahawk 6 open platforms.
Read deep-dive → Silicon · Access to CoreQumran3 Routers
Qumran3U at 1.2 Tb/s, Qumran3A at 2.4 Tb/s and Qumran3D at 25.6 Tb/s, and the four open routers UfiSpace and Edgecore build on them. OcNOS support is planned.
Read deep-dive → AI Fabric · Load balancingDynamic Load Balancing (DLB)
Flowlet rebinding using live ASIC queue-depth telemetry. Closes the ECMP hash-collision gap on GPU collectives.
Read deep-dive → AI Fabric · OcNOS 7.1Global Load Balancing (GLB)
Fabric-wide load balancing: end-to-end path scoring across leaf, spine, and super-spine for clusters of up to 16,384 GPUs. Coming in OcNOS 7.1.
Read deep-dive → AI Fabric · Next-GenUltra Ethernet (UEC)
UEC 1.0 explained: packet spray, multi-path RDMA, out-of-order delivery, modern congestion control. An open, standards-based alternative to InfiniBand.
Read deep-dive → AI Fabric · Reference DesignsAI Fabric Topologies
Rail-only and rail-optimized fabrics in two and three tiers. Sized in port counts for 256, 4k, and 16k GPU clusters on TH5.
Read deep-dive → AI Fabric · Congestion ControlDCQCN: RDMA Congestion Control
WRED ECN marking, CNP feedback, and quantized rate control. xCCL-class defaults out of the box; every threshold YANG-modeled for tuning.
Read deep-dive → AI Fabric · Deadlock recoveryPFC Deadlock Detection & Recovery
Per-port, per-priority detection timer that finds paused-queue cycles and auto-drains the affected queue before training jobs hang.
Read deep-dive → DC Fabric · Active/ActiveEVPN Multi-Homing (ESI-LAG)
RFC 7432 standards-based Active/Active server attachment. Designated forwarder, aliasing, mass-withdraw, no proprietary MLAG ICL.
Read deep-dive → DC Fabric · Inter-DC400G Coherent DCI (ZR / ZR+)
400G and 800G ZR and OpenZR+ pluggables direct in the spine. Transponder-free metro and long-haul DCI, with EVPN inter-DC for tenant extension where the platform supports it.
Read deep-dive → AI Fabric · Decision GuideInfiniBand vs Ethernet for AI
Workload-specific decision guide. Where modern Ethernet (RoCEv2 + DLB + UEC) closes the gap, where IB still wins, and how to pick.
Read deep-dive → Observability · gNMIStreaming Telemetry
gNMI Subscribe over gRPC, OpenConfig YANG, dial-out collectors, and integrations with Telegraf, Prometheus, and Grafana.
Read deep-dive → Timing · 5GPTP & SyncE
IEEE 1588v2 PTP, G.8273.2 Class C, SyncE G.8261/G.8262, boundary clock, transparent clock, hybrid mode, 30 ns per-hop time error.
Read deep-dive → Transport · RSVP-TEMPLS Traffic Engineering
RSVP-TE LSPs, CSPF path computation, Fast Reroute link/node protection, DS-TE class-types, auto-bandwidth.
Read deep-dive → Forwarding · SP + DCQoS and Traffic Management
DiffServ classification, srTCM and trTCM policing, per-queue and per-port shaping, WRED, weighted and strict-priority scheduling, three-level hierarchical queuing.
Read deep-dive → Operations · SLATWAMP & Service Assurance
TWAMP (RFC 5357) active measurement of IP delay, jitter and loss over MPLS and SR-MPLS L3VPN, and over SRv6 L3VPN on selected routers; Y.1564 service activation testing; Y.1731 Ethernet OAM with CFM, DMM/DMR delay and LMM/SLM loss.
Read deep-dive → Automation · Day-0NETCONF, YANG & ZTP
NETCONF over SSH/TLS, OpenConfig YANG, candidate datastore with commit/rollback, Zero-Touch Provisioning, Ansible.
Read deep-dive → Interactive toolAI Fabric Design Suite
Size a non-blocking leaf-spine fabric for your GPU cluster: leaf and spine switch counts, fabric links, and oversubscription on OcNOS-DC.
Open the tool → Data reportOcNOS Feature Coverage Report
About 7 in 10 OcNOS features implement a public standard, documented per platform on 44 qualified open platforms and 17 Broadcom ASICs.
View the report → PrimerOpen Networking
The open-networking vendor landscape and network operating system options, and where IP Infusion OcNOS fits.
Read the primer → ComparisonOcNOS vs Cisco
How OcNOS compares to Cisco IOS-XR and NX-OS: open hardware sourcing, silicon, licensing, standards parity, and support.
Compare → ComparisonOcNOS vs SONiC
Commercial supported NOS vs community open-source NOS for the data center, both on open merchant silicon: support model, integration burden, EVPN-VXLAN, and AI fabric.
Compare → ComparisonOcNOS vs Juniper
Service-provider migration from Juniper MX/QFX/ACX/PTX: platform mapping, Junos-to-OcNOS CLI, and a phased path.
Compare →Want a topic covered?
Tell us which protocol, architecture, or trade-off you'd like a deep-dive on. We write what operators ask for.
Design the whole AI fabric with OcNOS
From the business case to the port-count maths, pick up wherever you are in the build.