Distributed Scheduling for Buffered Crossbar Switches
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current packet switches face challenges in achieving high throughput and scalability due to complex centralized algorithms, particularly in handling non-uniform traffic patterns, which limits their effectiveness in large-scale high-speed switching systems.
Innovation Solution
A distributed scheduling method for buffered crossbar switches that stabilizes the system under any admissible Bernoulli traffic matrix by merging an initial priority schedule with a Hamiltonian walk schedule, allowing each input and output to make scheduling decisions based on local queue information and previous time slot schedules without message passing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If centralized scheduling algorithms (such as maximum weight matching) are used to achieve high throughput, then throughput is improved, but algorithm complexity increases significantly
Solution Approach 1:
The centralized scheduling function is segmented and distributed to individual input ports. Each input port independently performs scheduling decisions based on local queue information and crosspoint buffer states, eliminating the need for complex centralized matching algorithms while maintaining high throughput through parallel distributed operations
Solution Approach 2:
Each input port autonomously makes scheduling decisions without requiring centralized control. The input ports self-organize the scheduling process by independently selecting crosspoint buffers based on local information, achieving both simplicity and high throughput performance
2Device complexity
If buffered crossbar architecture with distributed scheduling is used, then device complexity is reduced, but achieving 100% throughput under non-uniform traffic becomes difficult
Solution Approach 1:
Crosspoint buffers are pre-configured and made available before scheduling decisions are made. Each input port can immediately select from available crosspoint buffers based on current queue states, enabling fast distributed scheduling decisions that achieve 100% throughput under non-uniform traffic patterns without complex algorithms
Solution Approach 2:
The scheduling approach changes from centralized matching algorithms to a parameter-based selection method where each input port selects crosspoint buffers based on simple local parameters (queue occupancy and buffer availability). This parameter-based distributed scheduling achieves both simplicity and 100% throughput under arbitrary traffic patterns
3Productivity
If centralized algorithms are used to handle non-uniform traffic patterns, then throughput is maintained, but scalability to large-scale switches is limited
Solution Approach 1:
The scheduling function is segmented across N input ports, allowing the system to scale to large N without increasing per-node complexity. Each input port performs identical simple scheduling operations independently, enabling linear scalability while maintaining 100% throughput under non-uniform traffic through parallel distributed execution
Data Source
AI summary
Scheduling methods and apparatus are provided for buffered crossbar switches with a crosspoint buffer size as small as one and no speedup. An exemplary distributed scheduling process achieves 100% throughput for any admissible Bernoulli arrival traffic. Simulation results also showed that this distributed scheduling process can provide very good delay performance for different traffic patterns. The simulation results also showed that packet delay is very weakly dependent on the switch size, which implies that the exemplary distributed scheduling process can scale with the number of switch ports.