Reinforcement Learning Load Balancer for Database Scalability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Database management systems struggle to simultaneously improve scalability and guarantee user-specified service level objectives, as they often rely on single storage backend databases, which limits flexibility and efficiency in handling varying traffic and performance requirements.
Innovation Solution
Implementing a hierarchy of backend databases with heterogeneous computing resources and using reinforcement learning to train a load balancer to select the optimal database based on speed and storage size, while also filtering requests with a cutoff frequency to optimize bandwidth usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single storage backend database is used, then system simplicity is maintained, but system scalability is limited
Solution Approach 1:
The patent divides the backend database into multiple storage backends with different characteristics (speed, size, cost) organized in a hierarchy. This segmentation allows the system to scale by selecting appropriate backends for different data types and access patterns, resolving the contradiction between scalability and complexity.
Solution Approach 2:
The patent introduces a hierarchical dimension to the backend database structure, organizing storage backends across multiple levels (hot, warm, cold storage). This dimensional organization enables scalable system design while managing complexity through structured hierarchy rather than flat complexity.
2Adaptability or versatility
If multiple backend databases are used to improve scalability, then system scalability is improved, but guaranteeing service level objectives becomes difficult
Solution Approach 1:
The patent implements a feedback mechanism where the system monitors service level objective performance and dynamically adjusts request routing decisions. The reinforcement learning model learns from past performance data to optimize routing policies, ensuring SLO guarantees are maintained while leveraging multiple backends for scalability.
Solution Approach 2:
The patent introduces dynamic request routing that adapts to changing system conditions and backend performance. Instead of static routing rules, the system dynamically selects which backend to use based on real-time metrics and learned patterns, maintaining SLO compliance while maximizing scalability benefits.
3Productivity
If reinforcement learning is used to train the load balancer, then optimal database selection is achieved, but computational overhead increases
Solution Approach 1:
The patent performs reinforcement learning training in advance during off-peak periods or system initialization, preparing optimized routing policies before they are needed. This preliminary action allows the system to benefit from learned optimization without incurring heavy computational overhead during actual request processing, as the model inference is much lighter than training.
Data Source
AI summary
A computer-implemented method, system and computer program product for maximizing system scalability while guaranteeing enforcement of service level objectives. A request is received to access a backend database in a hierarchy of backend databases that includes heterogenous computing resources with a dynamic range of performance. Upon receiving the request, a reinforcement learning based filter determines if the request's frequency of access exceeds a cutoff frequency. If the received request is not filtered, but instead, is passed through the filter, then one of the backend databases in the hierarchy is selected. Such a selection is made by a load balancer that is trained using reinforcement learning to select the optimal backend database taking into consideration the storage size and speed of the backend databases as well as taking into consideration the user-specified service level objective to be met by the request to guarantee enforcement of such a service level objective.