Credit-Based Multicast Switching Architecture
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional packet switching systems face scalability issues due to the increasing complexity of the switch unit as the number of ports and QoS priorities increase, leading to implementation bottlenecks in scaling the capacity of switching apparatus.
Innovation Solution
The solution decentralizes unicast traffic scheduling from the switch unit to output units using a credit-based mechanism, allowing output units to co-schedule unicast and multicast traffic on a per-output-port basis, thereby reducing the burden on the switch unit and enabling scalable implementation without increasing its complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If the switch unit performs all traffic scheduling and fabric scheduling functions, then the switching apparatus achieves centralized control and simplified architecture, but the switch unit complexity increases significantly as the number of ports and QoS priorities increase
Solution Approach 1:
The patent divides the scheduling function into two independent stages: traffic scheduling (performed by output units based on credit availability) and fabric scheduling (performed by the switch unit). This segmentation allows the switch unit to focus only on fabric scheduling while output units handle traffic scheduling, thereby reducing switch unit complexity as ports and QoS priorities increase.
Solution Approach 2:
The patent introduces credit tokens as an intermediary mechanism between output units and the switch unit. Credit tokens carry scheduling information and QoS priority data, enabling output units to perform traffic scheduling autonomously without burdening the switch unit with complex scheduling logic for each port and priority level.
2Adaptability or versatility
If the number of output ports and QoS priorities is increased to enhance switching capacity, then the versatility and throughput of the switching apparatus improve, but the complexity of the switch unit implementation increases
Solution Approach 1:
By separating traffic scheduling from fabric scheduling, the patent enables the system to scale to more ports and QoS priorities without proportionally increasing switch unit complexity. Output units independently handle traffic scheduling for their respective ports, while the switch unit only manages fabric scheduling across the switching fabric.
Solution Approach 2:
Output units autonomously perform traffic scheduling based on credit token information, eliminating the need for the switch unit to directly manage scheduling decisions for each port and QoS priority. This self-service capability at the output unit level allows the system to scale versatility without linearly increasing switch unit complexity.
Data Source
AI summary
A system and method of switching packets and/or cells, which includes a switching apparatus having a plurality of input units that receive at least one packet to be transferred by the switching apparatus. A plurality of output units transfer the packet out of the switching apparatus. A switch unit transfers the packet from one of the input units to one of the output units. Each input unit includes at least one input queue that temporarily holds the packet to be transferred by the switching apparatus. Each input unit also includes a respective unicast credit count unit that allows the packet to be transferred out from the queue when a current unicast credit value determined by the unicast credit count unit is at least predetermined value. Each output unit includes at least one output queue that receives the packet as switched by the switch unit, and which is to be transferred out of the switching apparatus. Each output unit also includes a output port scheduler unit that generates credit tokens and that outputs credit messages based on the credit tokens generated. Multicast credit count units are provided for each QoS priority that a multicast packet can be set to, in which they maintain a current multicast credit value for the QoS priorities.