The front-end fabric refers to the network infrastructure interconnecting the different functions of the AI environment.
This fabric facilitates communication, compute resource sharing, and data transfer among the cluster's interconnected clusters.
The front-end fabric also offers end users the necessary access to the servers and applications part of the AI environment. It also supports connectivity for network management through in-band interfaces.
When integrated with the storage fabric, the front-end fabric facilitates access to the storage, leading to a comprehensive unified fabric.
The bandwidth link speed from the Dell PowerEdge XE9680 to the front-end fabric is 400GbE using NICs 9 and 10 (see Figure 8), and they connect to a Dell Z9432F-ON.
The front-end fabric is a converged fabric where storage, application, and in-band GPU cluster management connections take place at different speeds ranging between 400GbE - 25GbE.
The front-end fabric uses EVPN VXLAN and standard Layer 2 or Layer 3 to provide multiple services such as multitenancy access into the AI environment, or straight access onto storage services from the GPUs as needed.
These VLAN segments (storage, application, and in-band GPU management cluster) are then routed within the front-end fabric to reach outside the AI environment or stay isolated as a private AI environment sandbox.
The connections from the storage, application, and in-band cluster management to the Dell PowerSwitch are Layer 2 in most cases. The different services (storage, GPU application, and in-band cluster management) are assigned a respective VLAN as it egresses the devices.