Automation, Orchestration, and Operations
This domain tests how you build and run an AI fabric repeatably: a source of truth, template-driven configuration, programmatic interfaces, and telemetry-driven validation instead of CLI sessions on individual switches.
The Cisco toolchain
- Cisco Nexus Dashboard is the operations platform hosting NDFC (fabric controller), NDI (Insights, for analytics and validation), and NDO (Orchestrator, for multi-site policy).
- NDFC discovers fabrics, applies per-switch and per-fabric policy templates, and provisions new switches through POAP (DHCP plus image and config delivery).
- NDI provides pre-change and post-change validation, flow analytics, and anomaly detection; it is an assurance tool, not a configuration tool.
- Cisco Intersight manages UCS and Intersight Cloud Orchestrator workflows server-side.
- Interfaces: YANG models over NETCONF/RESTCONF, gNMI streaming telemetry, REST APIs, Ansible collections, and Terraform providers. Prefer streaming telemetry over polling.
Worked example: adding 16 leaves
- Define the fabric in NDFC with the correct NX-OS version and policy templates.
- Pre-stage DHCP and the POAP definition so new switches receive image and management IP on first boot.
- Let discovery add the switches; NDFC computes the expected configuration from the fabric policy.
- Before the change window, run an NDI pre-change analysis against the proposed policy to catch reachability, MTU, VLAN, and QoS deltas.
- Deploy, then re-run compliance to confirm the running config matches intent. Reasoning: the template is the source of truth, the device config is derived. Anything typed manually creates drift that the next deployment overwrites or breaks.
Operational practice
- Track a compatibility matrix between Nexus Dashboard release, NX-OS release, and the NDFC/NDI versions; mismatches block upgrade paths.
- Maintain an inventory of GPU nodes and their rail mapping so a cable change does not silently break rail alignment.
- Automate firmware and BIOS compliance through Intersight profiles rather than one-off upgrades.
Traps that cost marks
- Treating NDI as a configuration interface.
- Using manual CLI changes that later conflict with fabric policy.
- Ignoring release compatibility between Nexus Dashboard and NX-OS.
- Polling SNMP every few minutes when nanosecond-scale queue counters are needed.