SDN Traffic Engineering via Alpha-Fairness Vector Operations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional SDN traffic engineering techniques exhibit polynomial computation scalability, making them unsuitable for large networks, and are unable to dynamically adjust performance characteristics like throughput and quality of service, limiting their efficiency and flexibility.
Innovation Solution
The implementation of alpha-fairness resource allocation schemes in SDN traffic engineering optimization, allowing for near-linear scalability and flexible policy execution through massively parallel processing, enabling simultaneous optimization of different paths and links, and adjustment of an alpha-fairness parameter for varying fairness and throughput levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional optimization algorithms are used for SDN traffic engineering, then specific objectives like max-min fairness or maximum throughput can be achieved, but the computational complexity increases significantly as additional links are added to the network
Solution Approach 1:
The patent segments the network optimization problem into independent link-level computations. Each link computes its own flow rates independently based on local shadow prices, rather than solving a monolithic network-wide optimization problem. This segmentation reduces computational complexity from polynomial to near-linear scalability while maintaining optimization objectives through iterative coordination of shadow prices across links.
Solution Approach 2:
The patent introduces dynamic adjustment of shadow prices for each link based on flow rate changes. The shadow prices are updated iteratively using a dynamic formula that responds to local conditions, allowing the system to adapt to changing network states without requiring complete re-optimization. This dynamic approach enables flexible response to traffic variations while keeping computational overhead low.
2Reliability
If conventional optimization algorithms are used for SDN traffic engineering, then specific performance objectives can be achieved, but the algorithms are incapable of being dynamically adjusted to provide different performance characteristics
Solution Approach 1:
The patent changes the fundamental parameter from fixed optimization objectives to dynamic shadow prices. By adjusting shadow prices as control parameters, the system can dynamically shift performance characteristics between throughput optimization and fairness considerations. Different performance modes are achieved by modifying shadow price update rules rather than changing the entire optimization algorithm, providing versatile adaptive control.
Solution Approach 2:
The patent implements dynamic performance adjustment through iterative shadow price updates that respond to real-time network conditions. The system can transition between different performance characteristics by modifying the dynamics of shadow price computation, allowing flexible adaptation to varying traffic patterns and service requirements without reconfiguring the underlying optimization framework.
3Reliability
If conventional linear programming schemes are used for SDN traffic engineering, then optimization can be performed, but execution times are significantly longer compared to parallel vector operations
Solution Approach 1:
The patent segments the optimization computation into independent vector operations that can be executed in parallel across multiple links. Each link performs independent calculations of flow rates and shadow prices using vector operations, eliminating the sequential dependency inherent in conventional linear programming. This segmentation enables massive parallel processing while maintaining optimization accuracy through iterative convergence of shadow prices.
Solution Approach 2:
The patent substitutes the mechanical sequential processing of linear programming with parallel vector operations. By replacing the traditional iterative linear programming mechanism with vector-based computations that can be simultaneously executed across multiple processors, the system achieves near-linear scalability and significantly reduced execution times while preserving optimization correctness through mathematical equivalence.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Traffic engineering vector operations that are capable of being independently solved can provide near-linear scalability through the exploitation of massively parallel processing. Optimization can be performed simultaneously on different paths in a data plane, as well as on different links within the same path (or within the same set of paths). In some embodiments, the traffic engineering vector operations include an adjustable alpha-fairness variable that allows managers to achieve different levels of fairness/throughput. Hence, embodiment alpha fairness techniques provide flexible policy execution, while maintaining excellent scalability for large network implementations.