Hello,
Fast glean is an optimization of the glean process on the Nexus 7000/7700 platforms.
The purpose of the classic glean process is to resolve an L3-to-L2 mapping (ARP or ND) of the immediate next hop (either another router on the path, or the destination itself if directly connected) for a packet if that mapping does not exist yet. If the linecard hardware does not know this mapping when routing a packet, it will punt the packet to the supervisor. NX-OS will pick the packet up in the IP driver but then will subject it to the entire software path of input and output processing including routing in software before the packet finally hits the ARP or ND code for the specific egress interface and triggers the ARP/ND resolution. This duplicates the input processing and routing already done by the linecard and obviously taxes the supervisor more than needed.
The idea of fast glean is to simplify this process for directly attached destinations. Instead of punting the packet into the general software IP path, every directly attached IPv4/IPv6 network is first assigned its own unique L3 adjacency index. When a packet requiring an ARP/ND resolution is punted to the supervisor, then instead of the general software IP path, it is punted directly to the ARP/ND driver and has the unique L3 adjacency index stored in an internal header. This allows skipping the entire software IP input path processing, ACLs, PBR, routing, etc., since the forwarding decision has already been made by the hardware, and instead, the packet is passed directly to the ARP/ND driver with an indication what is the egress interface where the resolution and ultimate forwarding is required (derived from the unique L3 adjacency).
The downside of this approach is that every directly attached network consumes a unique L3 adjacency index which is a system resource and may eventually get depleted.
Please feel welcome to ask further!
Best regards,
Peter
