Online Learning Algorithm for Kubernetes Resource Quota Configuration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing Kubernetes platform configures computing resources based on a preset standard quota, leading to irrational resource allocation due to varying resource needs among different service types, and this configuration is lost during system 'cold start' scenarios.
Innovation Solution
Implementing an online learning algorithm to recommend tailored computing resource quotas for each service type, using quota optimization algorithms like FTRL and UCB, and incorporating CPU and GPU resources to ensure rational allocation and initial configuration even in 'cold start' states.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a preset standard computing resource quota is used for all services, then the configuration process is simple and fast, but the resource allocation becomes irrational and inefficient for different service types
Solution Approach 1:
The patent applies local quality by transitioning from a uniform resource quota applied to all services to service-type-specific quotas. The system determines different computing resource quotas based on the specific service type (e.g., batch processing services vs. online transaction services), ensuring each service receives appropriately tailored resources rather than a one-size-fits-all allocation.
Solution Approach 2:
The patent changes the parameter of resource quota from a fixed preset value to a dynamic value that varies according to service type. The computing resource quota is adjusted based on service characteristics, transforming the static parameter into a flexible one that adapts to different service requirements.
2Speed
If computing resource quota is stored in system memory, then quick access is possible, but the configuration is lost during system cold start
Solution Approach 1:
The patent applies preliminary action by pre-determining and storing computing resource quotas in persistent storage (hard disk) rather than volatile memory. The system performs the configuration determination in advance and saves it to a file, ensuring the configuration is ready and preserved before system cold start occurs, thus maintaining both quick access capability and configuration persistence.
Solution Approach 2:
The patent uses copying by creating a persistent copy of the computing resource quota configuration from volatile memory to persistent storage (hard disk). This copy ensures that the configuration data survives system restarts and cold starts, allowing the system to retrieve the configured quotas after reboot without losing the allocation settings.
3Device complexity
If fixed resource configuration is used for all services, then system complexity is reduced, but resource waste or insufficiency occurs for different service types
Solution Approach 1:
The patent reduces resource waste by applying local quality - allocating computing resources according to the specific needs of each service type rather than using a uniform allocation. The system determines appropriate quotas based on service characteristics (batch processing vs. online transaction services), ensuring resources are neither wasted nor insufficient for any service category.
Data Source
AI summary
A method, system and apparatus for configuring a computing resource of a service are provided, which relate to the field of resource configuration. The method includes that: an online learning algorithm that is used for recommending, to each service on a platform, a computing resource quota adapted to a service type of the service is preset (S1); when a new service is launched on the platform, a target computing resource quota adapted to the service type of the new service is recommended according to the online learning algorithm (S2); and initial configuration is performed for a computing resource of the new service according to the target computing resource quota (S3). By means of the method, the computing resource quota adapted to the service type of a newly launched service on the platform may be recommended to the newly launched service according to the online learning algorithm, so as to make computing resource configuration of the service more rational. In addition, the method is independent of a standard computing resource quota stored in a system memory, and therefore may realize initial configuration on the resource of the service even in a “cold start” state.