Predictive Network Bandwidth Scheduling for Cloud Auto-Scaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Scheduling network resources in cloud computing is challenging due to the need to consider multiple physical network links and switches in hierarchical topologies, making it difficult to allocate bandwidth efficiently, especially when adding or removing virtual endpoints for auto-scaling use cases, leading to resource wastage and congestion.
Innovation Solution
A method and apparatus that utilize a pre-trained model for predicting network bandwidth and usage patterns based on local and central training, enabling the resource scheduler to allocate resource consumers optimally onto endpoints and network levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If network resources are scheduled considering hierarchical topologies and multiple physical network links, then network resource allocation accuracy is improved, but scheduling complexity increases
Solution Approach 1:
The patent segments the network scheduling problem into two distinct phases: an offline training phase where a machine learning model learns optimal scheduling patterns from historical data, and an online inference phase where the pre-trained model quickly predicts resource allocations. This segmentation transforms the complex real-time scheduling task into a manageable prediction task, maintaining high allocation accuracy while reducing online scheduling complexity.
Solution Approach 2:
The patent performs preliminary action by training the machine learning model offline using historical network traffic data and topology information before actual scheduling operations. This pre-training process captures complex network patterns and relationships in advance, enabling the system to make accurate scheduling decisions during runtime without having to compute complex allocations in real-time.
2Productivity
If bandwidth is allocated based on predicted network usage patterns, then network resource utilization is improved, but prediction accuracy requirements increase
Solution Approach 1:
The patent implements feedback by using actual network traffic measurements and utilization data to continuously refine and retrain the machine learning model. The system monitors predicted versus actual bandwidth usage patterns and uses this feedback to improve prediction accuracy over time, creating a closed-loop system that adapts to changing network conditions and improves resource utilization progressively.
Solution Approach 2:
The system performs preliminary bandwidth allocation based on predicted usage patterns before actual traffic flows occur. By using the pre-trained model to forecast bandwidth requirements in advance, the system can pre-allocate resources appropriately, improving utilization while the actual usage data later refines the predictions for future allocations.
3Object-generated harmful factors
If resource scheduler allocates bandwidth at multiple network levels, then network congestion is reduced, but allocation computation time increases
Solution Approach 1:
The patent segments the multi-level bandwidth allocation problem by having the machine learning model predict allocations at different network hierarchy levels (core, aggregation, access) simultaneously during the offline training phase. The pre-trained model then applies these segmented allocation decisions at appropriate network levels during online operation, reducing congestion throughout the hierarchy without requiring sequential computation at each level during runtime.
Data Source
AI summary
Provided are an apparatus and a method for scheduling network resources. In response to a resource request for a resource consumer, the resource scheduler obtains resource prediction including the predicted network bandwidth and the predicted network usage pattern, and schedules bandwidths for the resource consumer based on resource requirement information, the resource prediction and resource availability. The network bandwidths for the resource consumer are allocated based on the predicted network bandwidth and the predicted network usage pattern, network bandwidths can be allocated based on demand and supply, and can thus be fully utilized, that is, network bandwidth utilizations at the endpoint and the levels of networks can be improved.