Manage Your DC Fabric from Day 0 to Day N
Purpose-built fabric management platform for SONiC-based networks - covering design, provisioning, lifecycle automation, telemetry, and change governance in one unified system. Now with dedicated editions for AI factories, data centers, and enterprise.
One Platform. Design. Deploy. Operate.
Most teams stitch together disconnected tools for topology, config management, telemetry, and change control - then spend more time maintaining the glue than running the fabric. NetFabric replaces that entire stack with a single intent-driven platform.
Auto-discover your entire fabric in minutes with LLDP-based discovery across multi-tier Clos topologies. Vendor-agnostic support for Edgecore, Celestica, Dell, and all major ODMs. Intent-driven inventory tracking with hardware lifecycle flagging.
Template-driven config generation with Batfish pre-deployment validation. GitOps-native change pipeline with approval gates, change windows, and full audit trails. Every push is atomic with auto-rollback - config replace, never merge.
Sub-second gNMI streaming metrics exported to Grafana and InfluxDB with pre-built dashboards. Cross-fabric RCA engine isolates root cause in seconds. Proactive drift alerts notify you before incidents, not after.
Ask questions in plain English across topology, configuration, and live state. Context-aware validation suggests fixes before you commit. Guided workflows with built-in approval gates - AI assists, operators decide.
The Right NetFabric for Every Network
Same codebase. Same UI. Same renderer. Pick the edition that matches your deployment. Combine editions as you scale.
An AI factory isn’t one network - it’s four: a lossless Ethernet frontend, an InfiniBand scale-out backend, an intra-rack GPU interconnect, and an out-of-band management plane. Each has its own switching layer, its own failure domains, and often its own vendor tooling. The result: multiple consoles, spreadsheet-driven IP plans, and operational overhead that grows faster than compute.
Orion collapses that into a single intent-driven control plane. Operators describe what they need - "give tenant T-042 a 36-GPU partition with isolated RoCE QoS" - and NetFabric atomically renders the right config across all four network planes. With validation before push and rollback on failure.
- Reference design catalog: GB300 NVL72, GB200, DGX SuperPOD - pick a rack type, enter count, deploy
- Unified tenant lifecycle across IB PKey + NVLink partition + RoCE VPC + OOB VLAN - atomically
- 4-source telemetry unification: gNMI, UFM REST, NMX REST, DCGM - single pane of glass
- AI assistant indexed on your operator runbooks with natural-language queries over live state
- Simulation-first: ContainerLab + ibsim + DOCA-on-x86 - validate 80-90% without hardware
- NCCL-aware flow analytics - correlate fabric events to training job performance
- Future-proof: GB400, Rubin, MI400X enter the catalog as YAML - no fork needed
- Inference & serving traffic
- GPUDirect Storage (NVMe-oF)
- VPC + tenant VNI traffic
- Lossless RoCE: PFC / ECN / DCQCN
- NCCL all-reduce / all-to-all
- Cross-rack GPU-to-GPU RDMA
- Tenant-isolated PKey traffic
- Microsecond tail latency
- 72 GPUs as one coherent domain
- TB/s-class per-GPU bandwidth
- Managed tenant partitions
- Copper backplane - no optics
- BMC / IPMI out-of-band
- Zero-touch config push
- Per-rack VLAN + IP plan
- Survives if production is down
Running a multi-tenant data center means managing hundreds of pods, thousands of switches, and complex EVPN-VXLAN overlays - often across multiple vendors. Cygnus gives you the complete DC fabric stack with tenant-isolated VPCs, gateway services, and full operational visibility.
- EVPN-VXLAN underlay + overlay with automated BGP
- VPC + V-Net abstraction - per-tenant L2/L3 isolation
- Gateway service: NAT, L4 load balancing, elastic IPs
- Multi-vendor: Arista, Cisco, Cumulus, all SONiC variants
- GitOps config pipeline with Batfish pre-validation
- gNMI streaming telemetry with Grafana dashboards
- Cross-fabric RCA - correlate events across pods
- Build pod from scratch or import brownfield
Enterprise networks are messy - multi-vendor, brownfield, accumulated over years of organic growth. Polaris meets you where you are. Discover what's already running, model the intent, and start managing it like infrastructure-as-code. No rip-and-replace required.
- Zero-touch brownfield discovery - import your existing network
- Multi-vendor: Arista, Cisco, Cumulus, SONiC, and more
- L2/L3 routed access with WAN handoff templates
- Topology builder with reusable rack and site templates
- IPAM integration with pre-flight validation
- Digital twin - simulate changes before touching production
- Continuous drift detection with proactive alerts
- Build rack from scratch or discover existing infrastructure
Day 0 to Day N Lifecycle
Unified platform covering design, provisioning, and operations - not disconnected tools stitched together.
- Topology builder with Clos, leaf-spine, MLAG templates
- Reference design catalog - from enterprise 3-tier to GB300 NVL72
- IPAM integration - auto-allocate IPs, VLANs, VNIs, ASNs, PKeys
- Pre-flight validation via Batfish engine
- Digital twin simulation before any device is touched
- Zero-touch provisioning (ZTP) across all supported vendors
- Jinja2 / YANG config rendering - vendor-correct output
- Phased rollout with per-stage health gates
- Atomic config replace with auto-rollback on failure
- Automated acceptance testing before sign-off
- Git-based CI/CD pipeline for every config change
- Continuous drift detection - near-zero latency
- gNMI / gNOI streaming telemetry with Grafana dashboards
- Cross-fabric RCA with root-cause isolation
- Change windows, approval gates, and full audit trail
Built for Production. Layered for Extensibility.
Four clean layers from UI to device. Click each layer to explore.
Presentation Layer
Every operation - topology design, config commit, tenant creation, AI query - is accessible through all three interfaces. The React-based Web UI is the primary operator surface; the REST API and CLI serve automation and scripting workflows.
- React Web UI with real-time fabric visualization
- Full-featured REST API for infrastructure-as-code pipelines
- CLI for scripting, cron, and operator muscle-memory
- AI assistant embedded in all three surfaces
Orchestration API
The orchestration layer exposes two API surfaces: GraphQL for complex queries across the intent graph, and OpenAPI 3.0 for standard CRUD operations. Both are versioned, documented, and authenticated via the same RBAC system.
- GraphQL for topology traversal and cross-entity queries
- OpenAPI 3.0 with auto-generated SDKs
- Webhook callbacks for async provisioning workflows
- Rate limiting and RBAC at the API gateway
Fabric Intelligence Engine
The intelligence engine is where NetFabric earns its keep. It validates intent against resource pools, runs Batfish checks before any push, schedules phased rollouts, and correlates telemetry events into root-cause analysis.
- Batfish pre-deployment validation on every commit
- Resource pool engine - conflict detection on every change
- RCA engine correlates events across all fabric planes
- Scheduling engine for phased rollouts with health gates
- AIOps anomaly detection (Orion Phase 5)
Device Abstraction Layer
NetFabric speaks every protocol your devices understand. gNMI for modern SONiC telemetry, gNOI for operational RPCs, Netconf for YANG-aware config push, and SSH as the universal fallback. The abstraction layer ensures one intent renders to correct vendor config regardless of transport.
- gNMI streaming telemetry - sub-second metrics
- gNOI for operational commands (reboot, cert rotation)
- Netconf / RESTCONF for YANG-native devices
- SSH fallback for legacy and brownfield equipment
- Per-vendor device profiles - pluggable renderer architecture
NetFabric vs. the Alternatives
Purpose-built for SONiC-native operations - not a legacy SDN controller retrofitted for open networking.
| Capability | NetFabric PalC | Open-Source DIY |
|---|---|---|
| SONiC-native support | ✓ | ~ Partial |
| Multi-vendor ODM hardware | ✓ | ✓ |
| GitOps change pipeline | ✓ | ~ Manual assembly |
| Pre-deployment Batfish validation | ✓ | ✓ DIY integration |
| Day 0 digital twin / simulation | ✓ | ~ ContainerLab only |
| gNMI streaming telemetry | ✓ | ~ Partial |
| Cross-fabric RCA engine | ✓ | ✕ |
| AI-assisted operator UX | ✓ | ✕ |
| 4-fabric AI factory orchestration | ✓ Orion | ✕ |
| GPU rack reference design catalog | ✓ Orion | ✕ |
| Source-inspectable / escrow rights | ✓ | ✓ |
| Zero vendor lock-in | ✓ | ✓ |
| Professional 24/7 support & SLA | ✓ | ✕ |
Built for How You Actually Deploy
Orchestrate 4 network planes across thousands of GPUs. GB300, DGX, custom hybrid racks. Tenant-isolated training and inference with NCCL-aware analytics.
Multi-tenant GPU partitioning with atomic lifecycle across IB PKey, NVLink, and RoCE VPC. Billing-grade isolation, zero noisy-neighbor bleed.
EVPN-VXLAN multi-tenant fabrics with VPC isolation, gateway services, and L4 load balancing. GitOps-driven operations at pod scale.
Operate customer fabrics from a single pane. Multi-vendor support, RBAC per customer, per-pool audit logs. One platform, many tenants.
Discover and manage multi-vendor campus networks. Brownfield-friendly with L2/L3 routed access, WAN handoff, and drift detection built in.
Move from proprietary to SONiC at your own pace. Import existing topology, validate in simulation, cut over with confidence.
Let's Build Your Next-Gen Fabric Together
Whether you're orchestrating a multi-thousand-GPU AI factory or managing an enterprise campus, there's a NetFabric edition built for your scale. Early access customers get:
- Priority access to the platform before GA
- Dedicated onboarding with a PalC SONiC engineer
- Influence the product roadmap directly
- Early access pricing locked in at registration
Request Early Access
ODM and Other PARTNERS
Trusted by Industry Leaders
Next steps
Exploring fabric management for SONiC-based data centers?
Talk to the PalC team about NetFabric for Day 0 design, Day 1 provisioning, and Day N operations.
