Microservice Scaling via Reinforcement Learning Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current auto-scaling approaches for microservice-based applications in cloud computing scale resources locally, leading to delayed response times and inefficient resource allocation due to the lack of consideration for scaling delays across interconnected microservices.
Innovation Solution
A method and device utilizing a Reinforcement Learning (RL) model that takes into account the workload, current and historical resource allocations, and feedback signals to proactively scale microservices in a service mesh, minimizing resource allocation delays and optimizing resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If local auto-scaling control loops are used in each microservice, then resource allocation can be made independently and simply, but response time increases and scaling speed decreases due to lack of coordination across the service mesh
Solution Approach 1:
The patent merges multiple local control loops into a single centralized control loop that manages resource allocation across the entire service mesh. This centralized controller coordinates scaling decisions globally, eliminating the delays caused by sequential local scaling while maintaining manageable complexity through unified control logic.
Solution Approach 2:
The centralized control loop proactively anticipates workload changes and initiates resource allocation before performance degradation occurs. By monitoring workload patterns and predicting future demands, the system performs scaling actions in advance, reducing actual response time while maintaining simple local microservice implementations.
2Ease of manufacture
If local auto-scaling is implemented in each microservice, then implementation is simple and decentralized, but resource allocation efficiency decreases leading to high resource usage
Solution Approach 1:
The system implements a centralized feedback mechanism where the control loop continuously monitors workload across all microservices and adjusts resource allocation accordingly. This global feedback loop optimizes resource distribution efficiently, preventing both over-provisioning and under-provisioning, while maintaining simple local microservice code that doesn't require complex scaling logic.
3Device complexity
If reactive threshold-based scaling rules are used, then scaling decisions are simple to implement, but response time increases due to delayed detection of workload changes
Solution Approach 1:
The centralized control loop proactively monitors workload patterns and predicts future demands before threshold violations occur. By performing preliminary analysis of workload trends and initiating scaling actions in advance, the system reduces actual response time while maintaining simple threshold-based decision rules for scaling triggers.
Solution Approach 2:
The system implements continuous feedback monitoring of workload metrics across the service mesh, enabling the centralized controller to detect changes immediately and respond proactively. This real-time feedback mechanism maintains simple scaling rules while dramatically improving detection and response speed compared to periodic or event-driven local monitoring.
Data Source
AI summary
A method and a device for scaling microservices in a service mesh using reinforcement learning with a feedback signal. The reinforcement learning model uses information representing an input workload of a microservice chain and current and historical resource allocations of the service mesh, a reward, and the feedback signal to obtain an optimized resource allocation for the workload as an output of the RL model.