Multicast Packet Replication via Mask Division
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In network devices, the binary replication tree scheme for multicast operations leads to oversubscription of processing capacity and bandwidth, and updating multicast topology is time-consuming and prone to inconsistencies due to the need for static tree data structures.
Innovation Solution
The method involves creating a mask with addresses of egress PFEs, dividing it into two portions, and generating two copies of a packet, with each portion of the mask included in a separate copy, allowing each PFE to forward the packet to the next set of egress PFEs, thereby avoiding the need for static tree data structures and enabling quicker adjustments to multicast topology changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all necessary replications of the packet are performed on an ingress PFE, then every participating PFE receives the packet, but the ingress PFE's processing capacity and bandwidth are oversubscribed
Solution Approach 1:
The patent segments the packet replication process by distributing replication responsibilities across multiple PFEs in the fabric rather than concentrating all replication at the ingress PFE. Each PFE that receives a multicast packet performs local replication to its downstream PFEs, dividing the overall replication workload and preventing ingress PFE oversubscription while ensuring complete packet delivery to all participating PFEs.
2Productivity
If a binary replication tree scheme is used to reduce ingress PFE load, then processing capacity is improved, but multicast topology updates become time-consuming and prone to inconsistencies
Solution Approach 1:
The patent implements dynamic multicast topology management where the replication tree is not statically configured but dynamically adapted as packets traverse the fabric. Each PFE independently determines its replication actions based on real-time multicast group membership information, allowing the topology to automatically adjust to changes without requiring coordinated updates across all PFEs, thereby reducing update time and maintaining consistency.
3Device complexity
If static tree data structures are used in the binary replication tree scheme, then packet replication is organized, but the system is slow to adapt to multicast topology changes
Solution Approach 1:
The patent inverts the traditional approach by not pre-configuring static replication trees at each PFE. Instead, each PFE dynamically determines its replication behavior based on multicast group membership information received with packets or stored in local tables. This inversion from static preconfiguration to dynamic per-packet decision-making enables rapid adaptation to topology changes while maintaining organized replication through consistent decision logic at each PFE.
Data Source
AI summary
A network device receives a packet with a multicast nexthop identifier, and creates a mask that includes addresses of egress packet forwarding engines, of the network device, to which to provide the packet. The network device divides the mask into two portions, generates two copies of the packet, provides a first portion of the mask in a first copy of the packet, and provides a second portion of the mask in a second copy of the packet. The network device also forwards the first copy of the packet to an address of a first egress packet forwarding engine provided in the first portion of the mask, and forwards the second copy of the packet to an address of a second egress packet forwarding engine provided in the second portion of the mask.