A city GPU fabric’s Kubernetes AI namespace cannot schedule RDMA workloads yet. What is the deployment gate?
Select an answer to reveal the explanation.
Short Explanation
No operator, no RDMA schedule. Treat NVIDIA Network Operator deployment as the gate before the city GPU fabric’s AI namespace expects InfiniBand-attached pods.
Full Explanation
When a city GPU fabric Kubernetes AI namespace cannot schedule RDMA workloads yet, the deployment gate is to deploy the NVIDIA Network Operator before expecting RDMA/InfiniBand scheduling. Without the operator and its device/network plumbing, the scheduler and runtime lack the RDMA resource advertisements and attachment definitions those AI workloads need. Treating Network Operator deployment as a hard gate prevents false RDMA-ready declarations that waste municipal GPU cycles on pods that can never attach fabric devices.
Install the NVIDIA GPU Operator to provision InfiniBand drivers for RDMA scheduling fails because the GPU Operator manages GPU driver lifecycle and device plugins, not the InfiniBand fabric resource advertisements, network attachment definitions, or RDMA scheduling gate provided by the Network Operator. Configure Multus secondary interfaces for InfiniBand attachments using host-device plugins in the namespace fails as a partial workaround: it may expose interfaces but does not deploy the operator that advertises fabric capacity and gates RDMA scheduling for AI workloads. Advertise InfiniBand capacity with the Kubernetes Gateway API controller in the namespace fails because Gateway API manages service traffic routing, not the Network Operator deployment, RDMA device plugins, or attachment definitions needed to schedule fabric-attached pods.
Exam caveat: Deploying the Network Operator is the gate for expecting RDMA scheduling, but nodes must also have working InfiniBand HCAs and drivers. Operator install on Ethernet-only nodes will not invent RDMA hardware.
Operational check: Block production RDMA Job admission in the city AI namespace until Network Operator is deployed and verified, then enable a canary RDMA workload. Keep a checklist item for Network Operator deployed and healthy as a release gate for fabric-attached AI namespaces.