I am looking to do some work on a VNF to leverage hardware offloading features of the connectX-6 and ConnectX-7 cards when it comes to RSS hashing of MPLS packets. what would be the proper documentation to look into as to how to achieve this.
For ConnectX-6 and ConnectX-7, RSS and flow distribution are documented at the driver level. For Linux-based VNFs (DPDK or kernel networking), the main reference is the MLNX_OFED / MLNX_EN “RSS Support” chapter, which describes RSS hash functions, indirection tables and RSS flow steering for mlx5-based adapters:
For MPLS-encapsulated traffic specifically, the DPDK mlx5 NIC guide describes how MPLS (including MPLS-over-UDP / MPLS-over-GRE) is exposed to `rte_flow` and which match/actions are supported on Mellanox/NVIDIA adapters:
https://doc.dpdk.org/guides/nics/mlx5.html
If you are using VMware ESXi, the NVIDIA Native Drivers for VMware ESXi Inbox Drivers Release Notes explicitly list RSS support for MPLS – Added support for RSS on MPLS encapsulated packets in the Supported Features/Capabilities table:
The community forum can only point to these public documents. For a concrete design or validation of an RSS/MPLS scheme in your VNF (for example, exact hash fields and flow layout for ConnectX-6/7 in your environment), please work with your NVIDIA sales / account team or NVIDIA Enterprise Support so they can review the requirements in detail.