User Story:

As an OpenShift administrator using SR-IOV with Intel E810 NICs for high-performance workloads (e.g., NFV/Telco), I want the SR-IOV Operator to automatically configure the underlying node and drivers so that I can create the maximum number of queues per VF (e.g., 16) without requiring manual, node-level intervention and deep driver knowledge.

Problem Description:

The ice driver for Intel E810 series NICs exhibits default behavior that is suboptimal for SR-IOV use cases on nodes with a high CPU count. The driver's algorithm prioritizes balancing a small number of queues across all CPUs, which leads it to reserve an excessive number of MSI-X interrupt vectors for the Physical Function (PF).

For example, on a NIC with 512 total vectors, the driver may reserve 432 vectors for the PF. This leaves an insufficient number of vectors (80) for the Virtual Functions (VFs), preventing administrators from creating the desired number of RSS queues. In a common scenario, attempting to create 16 VFs results in each VF only receiving 4 queues, which is insufficient for many high-throughput applications.

The current workaround requires two separate, manual configurations:

1. A Tuned profile to set the number of active PF channels at runtime (ethtool).

1. A devlink kernel argument to limit the driver's MSI-X reservation at boot time.

This process is complex, error-prone, and requires deep, hardware-specific knowledge that should be abstracted away by the SR-IOV Operator.

Business Impact:
This issue complicates and adds manual steps to the configuration of high-performance networking for NFV and other latency-sensitive workloads. Automating and unifying this configuration would reduce administrative overhead, minimize errors, and ensure that customers can fully utilize the capabilities of their SR-IOV hardware without extensive manual tuning.