Serverless Runtime Configuration Prediction via Machine Learning

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

In serverless computing environments, traditional resource allocation methods do not effectively predict usage patterns or optimize resource utilization, leading to inefficiencies such as cold starts and suboptimal scaling, which can result in performance issues and increased costs.

Innovation Solution

A computer-implemented method that collects and analyzes attributes associated with applications in a serverless computing cluster, using machine learning models to predict usage patterns and dynamically update serverless runtime configurations, thereby optimizing container management and resource allocation based on anticipated workload.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional resource allocation methods are used in serverless environments, then resource allocation is simple to implement, but usage patterns cannot be predicted effectively leading to cold starts and suboptimal scaling

Engineering Contradiction:

Improveusage pattern prediction accuracyVSAvoidresource allocation system complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system performs preliminary actions by collecting attributes and using machine learning models to predict usage patterns before actual workload arrives. This allows the system to pre-emptively allocate resources and maintain warm containers, avoiding cold starts when requests are anticipated based on predicted patterns.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The resource allocation system uses self-service by implementing autonomous machine learning models that automatically analyze collected attributes, predict usage patterns, and adjust resource allocation without manual intervention. The system serves itself by making intelligent decisions about container warming and resource scaling based on predicted demand.

Inventive Principle:

Principle #25Self-service

2Productivity

If resources are maintained in warm state to reduce cold start times, then performance improves, but resource costs increase

Engineering Contradiction:

Improverequest handling speedVSAvoidresource consumption

Core Design Contradiction:

ProductivityVSLoss of energy

Solution Approach 1:

The system applies dynamics by making resource allocation flexible and adaptive rather than static. Machine learning models continuously analyze usage patterns and dynamically adjust which containers are maintained in warm state, scaling resources up or down based on predicted demand. This allows the system to optimize between performance and resource consumption in real-time.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system changes parameters by using machine learning to predict when resource allocation parameters should be adjusted. Based on predicted usage patterns, the system dynamically modifies container warm-up timing, resource allocation levels, and scaling parameters to achieve optimal balance between reducing cold starts and minimizing resource consumption.

Inventive Principle:

Principle #35Parameter changes

3Adaptability or versatility

If machine learning models are used to predict usage patterns, then resource allocation is optimized, but system complexity increases

Engineering Contradiction:

Improveresource allocation adaptabilityVSAvoidprediction system complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system implements feedback by using machine learning models that continuously analyze collected attributes from the serverless environment and use this feedback to improve usage pattern predictions. The models learn from historical data and actual outcomes, adjusting their predictions to better anticipate workload demands and optimize resource allocation decisions.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The machine learning models serve as intermediaries between raw attribute data and resource allocation decisions. Rather than directly implementing complex allocation logic, the system uses ML models as mediators that process attribute data, predict usage patterns, and translate these predictions into actionable resource allocation decisions, simplifying the overall system architecture.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS11652688B2Predicting usage pattern of serverless environment via machine learning

Publication Date: 2023.05.16 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

An approach for updating one or more serverless runtime configurations based on predicting usage patterns of a serverless computing cluster is disclosed. The approach includes collecting one or more attributes associated with an application on a serverless computing cluster and a user of the application and analyzing the one or more collected attributes. Furthermore, the approach includes predicting one or more usage patterns based the analyzing and updating one or more serverless runtime configurations based on the one or more usage patterns.