MIOQ Switch Parallel Crossbar Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-speed switches face challenges in achieving both high switching capacity and quality-of-service (QoS) guarantees, with existing technologies like input-queued and combined input/output-queued switches requiring speedup and complex scheduling algorithms, making them impractical for high operation rates.
Innovation Solution
A multiple input/output-queued (MIOQ) switch with a multi-token-based arbiter for scheduling and a virtual FIFO queueing scheme to ensure packet order, allowing for a simple round-robin operation that matches the performance of an output-queued switch without speedup, using a (k,m)-dimensional crossbar fabric and iterative matching algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a CIOQ switch is used to achieve both high switching capacity and QoS guarantees, then the switching capacity is improved, but the device complexity increases due to speedup requirements and complex scheduling algorithms
Solution Approach 1:
The switch fabric is divided into multiple parallel crossbar switches (e.g., two 8×8 crossbars instead of one 16×16 crossbar). Each parallel crossbar handles a subset of input-output pairs, reducing the complexity of individual scheduling decisions while maintaining overall high switching capacity through coordinated operation of multiple smaller units.
Solution Approach 2:
The patent introduces a time dimension by implementing time-division multiplexing across parallel crossbars. Different crossbars operate at different time slots, effectively adding a temporal dimension to the spatial parallelism. This allows the system to achieve high switching capacity without requiring each individual crossbar to handle all inputs and outputs simultaneously, thereby reducing scheduling complexity.
2Device complexity
If a parallel switch is used to avoid speedup requirements, then the device complexity is reduced, but the productivity decreases due to dependence on complex centralized distribution algorithms
Solution Approach 1:
Each parallel crossbar is equipped with its own independent scheduling algorithm that makes local scheduling decisions based on local queue states. This self-service approach eliminates the need for a complex centralized controller, allowing each crossbar to operate autonomously at high speed while contributing to the overall system performance.
Solution Approach 2:
The patent implements dynamic load balancing and traffic distribution mechanisms that adapt to changing traffic patterns in real-time. The system dynamically adjusts which crossbar handles which traffic flows, optimizing performance without requiring complex centralized coordination. This dynamic behavior enables high operation rates while maintaining simplicity.
3Productivity
If memory access time is shortened to support higher speedup, then the switching capacity is improved, but the loss of time for scheduling decisions increases
Solution Approach 1:
The scheduling function is segmented and distributed across multiple parallel crossbars, each with its own simplified scheduler. This segmentation reduces the computational burden on any single scheduler, allowing scheduling decisions to be made quickly without requiring extremely fast memory access times. The overall switching capacity is maintained through the combined effect of multiple parallel scheduling operations.
Data Source
AI summary
A multiple input/output-queued(MIOQ) switch is presented. This MIOQ switch comprises a (k,m)-dimensional crossbar fabric having k ingress lines and m egress lines; N input buffers at each input; M output buffers at each output; N×k interconnection networks each of which is able to move a packet from each input to one of the k egress lines; and m×M interconnection networks each of which is able to move the arrived packet at output from the m egress lines to one of the M(wherein M is m or N) output buffers. The multiple input/output-queued(MIOQ) switch according to the present invention requires no “speed-up”. Performance matches the performance of an output-queued switch.