Load Balancer Feedback Control for Dynamic Traffic Class Shifting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing load balancing techniques in cloud-delivered Secure Access Service Edge (SASE) products, such as cloud-delivered VPNs, fail to ensure consistent utilization of backend server nodes due to insufficient load balancing criteria, especially when traffic levels are inconsistent, and lack the ability to automatically adjust 'pinning' behavior or upgrade backend processes efficiently.
Innovation Solution
Implementing a feedback control loop using backend server nodes to provide additional metrics to load balancers, allowing for dynamic load balancing by sending congestion notifications and adjusting traffic based on predicted and actual load capacities, and dynamically upgrading data flows to higher traffic classes when resources are available.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If load balancers use traditional load balancing algorithms to minimize delay, then traffic handling performance is improved, but accuracy in ensuring full server node utilization deteriorates
Solution Approach 1:
The patent implements a feedback control loop where load balancers continuously monitor actual server node utilization and adjust traffic distribution accordingly. The system receives feedback about server capacity and dynamically modifies load balancing decisions to achieve both fast traffic handling and accurate utilization targeting, resolving the contradiction between speed and precision.
Solution Approach 2:
The load balancing system transitions from static algorithms to dynamic adjustment mechanisms. The patent enables load balancers to adaptively change traffic distribution in real-time based on current server conditions, allowing the system to maintain both high performance and accurate utilization across varying traffic patterns.
2Ease of operation
If ECMP routing is used to spread VPN traffic across backend nodes, then traffic distribution is simplified, but ability to implement pinning behavior and automatic adjustment deteriorates
Solution Approach 1:
The patent segments the load balancing function into multiple components: ECMP routing for basic traffic distribution and additional control mechanisms for pinning behavior. By dividing the functionality, the system maintains the simplicity of ECMP while adding specialized capabilities for automatic adjustment and node pinning through separate control planes.
Solution Approach 2:
The patent introduces an intermediary control mechanism between ECMP routing and backend nodes. This intermediary layer enables pinning behavior and automatic adjustment by mediating traffic flow decisions, allowing ECMP to continue handling general distribution while the intermediary provides sophisticated control for specific nodes or traffic patterns.
3Stability of the object's composition
If load balancers direct traffic to backend server nodes without dynamic adjustment, then system stability is maintained, but resource utilization consistency deteriorates
Solution Approach 1:
The patent implements feedback control loops that continuously monitor server node utilization and provide real-time adjustments to load balancers. This feedback mechanism maintains system stability by making gradual, controlled changes while ensuring consistent resource utilization across backend nodes, preventing both overload and underutilization.
Solution Approach 2:
The load balancing system enables backend server nodes to self-report their capacity and utilization status. Nodes automatically provide information about their current state, allowing the load balancer to make informed decisions without external intervention, thereby maintaining stability while optimizing utilization through decentralized self-management.
Data Source
AI summary
Techniques for dynamically load balancing traffic based on predicted and actual load capacities of data nodes are described herein. The techniques may include determining a predicted capacity of a data node of a network during a period of time. The data node may be associated with a first traffic class. The techniques may also include determining an actual capacity of the data node during the period of time, as well as determining that a difference between the actual capacity and the predicted capacity is greater than a threshold difference. Based at least in part on the difference, a number of data flows sent to the data node may be either increased or decreased. Additionally, or alternatively, a data flow associated with a second traffic class may be redirected to the data node during the period of time to be handled according to the first traffic class.