Connector software switch in Low CBF

Connector Deployment Strategy

Low CBF connector is in essence the control block of our software switch, be it NVidia, Broadcom, CISCO, or any emerging software switches.

As such, the connector is leveraging API/SDK providing by the vendor. Usually, those SDK are not open source and would require to be deployed directly on the switch of choice.

In order to interact with those SDK, we have chosen to build our own gRPC service on top of the proprietary SDK. Doing so, we must have this gRPC service directly deployed on the the switch, whether on baremetal or pod but definitely on the switch itself.

Once the gRPC service deployed, the tango side of the equation is becoming quite straightforward as the tango device simply use the gRPC client associated with the deployed service.

As a result we have a clear separation of sofware as shown in the diagram below:

  • the ska_low_cbf_nvidia_connector package:

    • Deployment location: on the switch

    • Function: implements routing and low level monitoring using SDK, then expose those functions via gRPC interface.

  • the ska_low_cbf_neoconnector package:

    • Deployment location: in the k8s cluster

    • Function: tango control of the switch leveraging gRPC communication with the switch

Connector Sofwtare Organisation