Multi-Stage Switch Fabric Arbitration for Algorithm Complexity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing switch fabric systems face complexity and area challenges due to complex algorithms and increased buffer requirements, making it difficult to implement efficiently, especially with a large number of input and output ports, leading to low performance.
Innovation Solution
A switch fabric system utilizing two arbiters for scheduling data between crossbar units, reducing algorithm complexity and area requirements by breaking down the N×N scheduling into two lower complexity algorithms, O(L×N) and O(M×1), and implementing a multi-stage buffered crossbar structure with cascaded CU units.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If a bufferless crossbar structure is used, then the area is reduced, but the algorithm complexity increases
Solution Approach 1:
The patent divides the N×N crossbar into multiple smaller crossbar units (CUs), each handling a subset of input/output ports. This segmentation reduces the algorithm complexity from O(N²) for a single large crossbar to O(M×N) for multiple smaller CUs, while also reducing the area required for each individual crossbar unit.
2Device complexity
If a buffered crossbar structure is used, then the algorithm complexity is reduced, but the buffer area increases
Solution Approach 1:
The patent distributes buffers across multiple smaller crossbar units rather than using a single large buffer. Each CU has its own buffer, reducing the peak buffer area requirement at any single location while maintaining the buffering functionality needed for low algorithm complexity.
3Adaptability or versatility
If the number of input and output ports increases, then the switching capability is improved, but the implementation difficulty increases
Solution Approach 1:
The patent implements a modular architecture where the large N×N switching fabric is divided into multiple smaller crossbar units. This modular approach makes implementation more manageable by breaking down the complex task of implementing a large switch into smaller, more tractable units that can be designed and manufactured separately.
Solution Approach 2:
The patent transitions from a single-level N×N crossbar to a multi-level hierarchical structure with multiple stages of crossbar units. This dimensional change from a flat single-layer architecture to a multi-layer hierarchical architecture reduces the complexity of individual units while maintaining overall switching capability.
4Device complexity
If the number of arbiters is reduced, then the device complexity is reduced, but the scheduling capability deteriorates
Solution Approach 1:
The patent distributes arbitration functions across multiple smaller arbiters, each responsible for a subset of ports, rather than using a single complex N×N arbiter. This segmentation reduces the complexity of each individual arbiter while collectively maintaining the scheduling capability needed for the entire switching fabric.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
Embodiments of the present invention provide a switch fabric system and a data switching method. The switch fabric system of the present invention includes: M first crossbar units CUs and N second CUs, where each first CU includes: L first input ports, a first arbiter, a first crossbar, and N first output ports, each second CU includes: M second input ports, a second arbiter, a second crossbar, and one second output port, and M×N first output ports of the M first CUs are respectively coupled to N×M second input ports of the N second CUs, where N first output ports of each first CU are respectively coupled to and in a one-to-one correspondence with one second input port of each second CU in the N second CUs, N = M × L, and M, N, and L are all positive integers. The embodiments of the present invention reduce complexity of an algorithm and an area of a crossbar, and can reduce difficulty in implementing an SF system and improve performance when there is a relatively large quantity of input ports and output ports.