TOR Switch Data Scheduling via Dynamic Path Load Balancing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data flow scheduling schemes in large-scale data center networks (DCNs) fail to implement real-time load balancing across multiple paths, leading to inefficient data transmission and severe load unbalancing due to high dynamics and heterogeneous network environments.
Innovation Solution
A data scheduling method for Top of Rack (TOR) switches that dynamically selects paths based on load conditions, using a combination of historical path optimization and random selection of candidate paths to ensure convergence and reduce storage and calculation overheads, with a focus on maintaining a preset quantity of paths to balance load and minimize detection delay.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If real-time load balancing is implemented across multiple paths, then network performance is improved, but detection delay and calculation overhead increase
Solution Approach 1:
The patent pre-calculates and stores load information for multiple paths before actual data transmission is needed. The controller proactively gathers path status information and maintains it in a ready state, so when path selection is required, the decision can be made immediately using pre-prepared data, thus reducing detection delay while maintaining real-time load balancing capability
Solution Approach 2:
The patent extracts only the necessary path selection function from the complex network control system. Instead of implementing full real-time monitoring and control of all network parameters, the system extracts and focuses on collecting and utilizing load information from multiple paths, simplifying the detection process while achieving the core goal of load balancing and improving network performance
2Productivity
If load information from multiple paths is collected and utilized, then load balancing is achieved, but storage and calculation overheads increase
Solution Approach 1:
The patent extracts only the essential load information needed for path selection from the complete set of network parameters. Instead of collecting and processing all possible network data, the system focuses on extracting and utilizing specifically the load information from multiple paths, reducing storage requirements and calculation overhead while maintaining effective load balancing capability
Solution Approach 2:
The patent implements partial action by collecting load information from a selected number of paths rather than all possible paths in the network. This partial collection approach provides sufficient information for effective load balancing decisions while significantly reducing the storage and calculation resources required compared to comprehensive monitoring of every possible path
Data Source
Figure 1
Figure 2
Figure 3
AI summary
This application provides a data scheduling method, applied to a first top of rack TOR switch in a data center network. The method includes: receiving data whose destination address is a second TOR switch; randomly selecting candidate paths, where the candidate paths include a historical path and a preset quantity of paths that are selected from the data center network and whose start ends are the first TOR switch and termination ends are the second TOR switch, and the historical path is a path that is selected in a historical data scheduling process and used to transmit historical data and whose start end is the first TOR switch and termination end is the second TOR switch; and when the candidate paths are selected, scheduling the data to a path whose load is not greater than a preset threshold in the candidate paths. Load of a path is used as a basis for data scheduling, so that a load change status of the DCN can be dynamically sensed. Therefore, load balancing of a plurality of paths in a large-scale DCN network can be implemented.