Digital Optical Monitoring (DOM)
Your AI fabric runs on light. Observe every photon of it.
Data Centers and AI Factories Idle on Bad Optics
Dorado delivers end-to-end AI and data center networking solutions spanning 10G to 1.6T, enabling reliable, high-performance connectivity across the modern data center. At these speeds, a single degraded transceiver or marginal cable can stall a distributed workload — leaving the most expensive silicon in the building waiting on light instead of processing data.
Optical links are often the least visible part of the infrastructure — invisible until utilization drops and performance suffers, with no easy way to identify which of thousands of links is responsible.
Cruz Optical Assurance makes the optical layer observable, correlated, and acceptance-tested — as one motion with the rest of the fabric.
Built for Rack Scale AI
Purpose-built for dense, high-radix optical fabrics on modern rack-scale GPU platform architectures — across the front-end, GPU back-end, and storage tiers.
72-GPU NVL-class Racks
8-GPU HGX-class Nodes
OCP Racks (ORv3/ORW
PCIe / Inference Racks
AMD MI355x/MI400x & Helios Architectures
Multi-GPU Racks.
What it does
Continuous Optical DOM
Live digital optical monitoring across every transceiver — Rx/Tx power, temperature, laser bias, per-lane health — normalized into one fabric-wide view so drift is caught before it becomes downtime.
Transceiver & Cabling Intelligence
Full inventory and reach validation — QSFP-DD 400G, OSFP 800G/1.6T, DAC, AOC, single- and multi-mode fiber — including LPO-ready linear pluggable optics that cut interconnect power by roughly 60%.
Liecycle, One Maintenance
Discovery, validation, and monitoring of the optical layer run as one lifecycle with the fabric — resell, deploy, and support — so a build arrives production-ready instead of project-ready.
Link -health + Acceptance Evidence
Optics DOM and link health fold into the fabric acceptance report alongside BGP/ECMP symmetry, RoCEv2 (PFC/ECN), and MTU — a signed, repeatable PDF, not a screenshot of green lights.
Fault →Workload Correlation
An optics or link fault resolves through the graph to the exact GPUs, jobs, and tenants it affects — blast-radius in a single traversal, so the on-call engineer knows the impact, not just the alarm.
Rail & Plane-Aware Resilience
Understands twin-planar / rail-optimized design, so a single transceiver or cable failure degrades to half-bandwidth rather than aborting a job — and flags when redundancy is gone.
How it works
STEP 1
Discover & Baseline
Inventory every transceiver and cable across the fabric, and set healthy baselines for power, temperature, and bias.
STEP 2
Monitor Continuously
Live DOM telemetry streams from every link, normalized into one fabric-wide view — drift shows up before it becomes downtime.
STEP 3
Alert & Correlate
A fault resolves through the graph to the exact GPUs, jobs, and tenants it touches, so on-call knows the impact, not just the alarm.