Data Center Congestion Control via Two-Level Bandwidth Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current congestion control mechanisms in data center networks are inadequate in managing congestion effectively, leading to increased end-to-end delay, queue length, and packet loss, especially due to the lack of perception among receive ends and the presence of multiple intermediate forwarding devices.
Innovation Solution
A congestion control method that involves determining weights for packet flows and allocating bandwidth based on these weights to ensure proper bandwidth allocation across different ports and destinations, allowing for flow-level scheduling and eliminating congestion by coordinating between forwarding devices and destination devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If active congestion control mechanism is used, then the last hop congestion is prevented, but multiple congestion points may still occur in the network due to lack of perception among receive ends
Solution Approach 1:
The patent introduces a centralized controller as an intermediary that coordinates bandwidth allocation across multiple forwarding devices. The controller receives bandwidth allocation requests from destination devices, calculates optimal bandwidth distribution considering network state, and distributes allocation decisions to relevant forwarding devices. This mediator approach resolves the contradiction by enabling coordinated congestion control across multiple network points without requiring direct perception or complex peer-to-peer coordination among receive ends.
Solution Approach 2:
The patent implements a feedback mechanism where destination devices monitor network congestion state and send bandwidth allocation requests to the centralized controller. The controller continuously receives network state information from forwarding devices, adjusts bandwidth allocation decisions based on this feedback, and updates allocations dynamically. This feedback loop enables the system to respond to changing network conditions and prevent multiple congestion points while maintaining manageable device complexity.
2Productivity
If bandwidth allocation is performed at forwarding device level, then flow-level scheduling is achieved, but notification to destination devices increases communication overhead
Solution Approach 1:
The patent merges multiple bandwidth allocation notifications into a single centralized control process. Instead of each forwarding device independently notifying its destination devices, the centralized controller consolidates bandwidth allocation decisions and sends unified notifications. This merging approach maintains flow-level scheduling efficiency while reducing redundant communication overhead, as the controller aggregates allocation information from multiple forwarding devices and communicates only essential updates to destination devices.
3Quantity of substance
If multiple forwarding devices are deployed to handle packet flows, then network capacity increases, but congestion points multiply
Solution Approach 1:
The centralized controller acts as a mediator that coordinates bandwidth allocation across multiple forwarding devices. It receives network state information from all forwarding devices, calculates optimal bandwidth distribution to prevent congestion, and distributes allocation decisions to appropriate devices. This intermediary coordination ensures that increased network capacity from multiple forwarding devices does not lead to proportional increase in congestion points, as the controller harmonizes their operations.
Solution Approach 2:
The patent dynamically changes bandwidth allocation parameters based on real-time network state. The centralized controller monitors congestion levels, traffic patterns, and device capacities, then adjusts bandwidth allocations to forwarding devices and destination devices accordingly. This dynamic parameter adjustment ensures that multiple forwarding devices operate efficiently without creating congestion points, as allocations are continuously optimized based on actual network conditions.
Data Source
AI summary
A congestion control processing method uses a two-level scheduling manner of a forwarding device and a destination device, where the network device of a data center network performs coarse-grained bandwidth allocation based on weights of flows destined for different destination devices. The network device allocates each flow a bandwidth that does not cause congestion, and notifies the destination device. The destination device performs fine-grained division, determines a maximum sending rate for each flow, and notifies a packet sending device of the maximum sending rate.