Split Architecture Network Congestion Control via Centralized Throttling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current congestion control techniques in networks are sub-optimal due to limited knowledge about network congestion status, leading to independent and often sub-optimal decision-making by nodes, resulting in inefficient resource utilization and unfairness among users.
Innovation Solution
A centralized controller in a split architecture network collects flow statistics, identifies bottleneck links, calculates throttling probabilities for entities sharing the target path, and transmits these probabilities to ingress switches to manage congestion dynamically, ensuring fair resource allocation and efficient bandwidth use.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If distributed congestion control methods are used where each node makes independent decisions, then device complexity is reduced and ease of operation is improved, but resource utilization efficiency deteriorates and fairness among users worsens
Solution Approach 1:
The patent introduces a centralized controller as an intermediary between network nodes and users. The controller collects congestion information from all nodes, performs global optimization calculations, and sends back rate adjustment decisions to individual nodes. This mediator approach resolves the contradiction by maintaining simple node operation while achieving efficient resource utilization through centralized coordination.
2Device complexity
If distributed congestion control methods are used where each node makes independent decisions, then device complexity is reduced, but resource utilization efficiency and fairness among users deteriorate
Solution Approach 1:
The centralized controller serves as an intermediary that assumes the computational complexity of congestion control. Individual nodes remain simple forwarding elements that report status to the controller and execute its decisions. This distribution of complexity resolves the contradiction by keeping node complexity low while achieving high resource utilization through the controller's global optimization capabilities.
3Ease of operation
If end-to-end congestion control mechanisms like TCP are used, then fairness among users with similar round-trip times is improved, but resource utilization efficiency deteriorates due to sub-optimal decisions from limited local knowledge
Solution Approach 1:
The patent implements a feedback mechanism where the centralized controller continuously collects congestion information from all network nodes, processes this global information to determine optimal rate allocations, and feeds back rate adjustment decisions to individual flows. This feedback loop resolves the contradiction by providing flows with global network state information, enabling both fairness and optimal resource utilization.
4Productivity
If centralized control is implemented to achieve network-wide optimization, then resource utilization efficiency and fairness are improved, but device complexity and control system complexity increase
Solution Approach 1:
The patent segments the control system into a centralized controller that performs global optimization and individual nodes that execute local actions. The controller handles the complex tasks of information aggregation, bottleneck identification, and rate calculation, while nodes perform simple tasks of monitoring and rate adjustment. This segmentation resolves the contradiction by concentrating complexity in a dedicated control element while keeping the network infrastructure relatively simple.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A controller performs network-wide congestion control in a split architecture network. The controller receives flow statistics from switches for each entity that communicates via the network. The controller chooses a most congested link as a bottleneck link based on a packet loss rate that is derived from the flow statistics and estimated for each entity on each link of the network. The controller identifies a target path in the network that passes through the bottleneck link and carries the most traffic, and calculates a throttling probability for each sharing entity that shares the target path based on the capacity of the bottleneck link, capacity of the first link of the target path, bandwidth consumption and packet loss rate of each sharing entity. The controller then transmits the throttling probability of each sharing entity to the ingress switch of the target path to reduce the congestion on the bottleneck link.