ML Admission Control for Cloud Resource Availability Prediction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Cloud provider networks face challenges in efficiently managing resource allocation and minimizing the number of insufficient capacity exceptions (ICEs) due to the complexity of scaling and maintaining rule-based systems for admission control, which can lead to suboptimal resource utilization and customer experience.

Innovation Solution

Implementing a machine learning-based admission control service that predicts optimal access control actions by training models on historical and simulated data to determine whether to approve or deny resource requests, thereby reducing the number of ICEs and improving resource pool protection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If rule-based systems are used for admission control, then resource allocation can be managed, but the system complexity increases and suboptimal resource utilization occurs

Engineering Contradiction:

Improveresource allocation efficiencyVSAvoidadmission control system complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent replaces rule-based mechanical admission control systems with a machine learning model that learns optimal admission decisions from historical data. The ML model substitutes complex manual rule configuration and evaluation with automated pattern recognition, reducing system complexity while improving resource allocation efficiency through data-driven decisions.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms the admission control system from using fixed rules to using dynamic parameters learned from historical data. The ML model processes multiple input features (resource availability, demand patterns, customer priorities) and generates adaptive admission decisions, allowing the system to optimize resource utilization based on changing conditions rather than static rules.

Inventive Principle:

Principle #35Parameter changes

2Reliability

If rule-based admission control is implemented, then resource management is possible, but customer experience deteriorates due to suboptimal decisions

Engineering Contradiction:

Improveresource pool protectionVSAvoidcustomer experience

Core Design Contradiction:

ReliabilityVSEase of operation

Solution Approach 1:

The patent implements a feedback mechanism where the ML model is trained on historical admission control data including outcomes and customer responses. This feedback loop allows the system to learn from past decisions and continuously improve admission strategies, balancing resource pool protection with customer satisfaction by adapting to actual system performance and customer needs.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent uses the ML model to make preliminary admission decisions before resources are fully allocated or rejected. By predicting likely outcomes and optimizing admission choices in advance based on historical patterns, the system can protect resource pools while providing better customer experience through more accurate and timely admission decisions.

Inventive Principle:

Principle #10Preliminary action

3Measurement precision

If manual intervention is used for admission decisions, then control accuracy can be maintained, but response time increases and scalability is limited

Engineering Contradiction:

Improveadmission decision accuracyVSAvoidadmission decision speed

Core Design Contradiction:

Measurement precisionVSSpeed

Solution Approach 1:

The patent implements self-service admission control where the ML model autonomously makes admission decisions without requiring manual human intervention. The model has been trained to achieve high decision accuracy by learning from historical data, and it automatically processes admission requests in real-time, providing both high speed and maintained accuracy through automated intelligent decision-making.

Inventive Principle:

Principle #25Self-service

4Productivity

If resource allocation is optimized, then resource pool protection improves, but the number of ICEs increases without proper prediction

Engineering Contradiction:

Improveresource utilization optimizationVSAvoidinsufficient capacity exception rate

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The patent applies the ML model to make preliminary predictions about resource availability and admission outcomes before actual allocation decisions are made. By forecasting likely ICE conditions based on historical patterns and current system state, the system can optimize resource utilization while proactively preventing excessive ICEs through informed admission control decisions.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11245640B1Systems, methods, and apparatuses for predicting availability of a resource

Publication Date: 2022.02.08 AMAZON TECH INC

AI summary

Techniques for predicting the availability of a resource are described. An exemplary method includes obtaining capacity data indicating an amount of capacity available in a cloud provider network to satisfy the request; generating, using a machine learning model that has been trained based at least in part on an output of an automated historical hindsight learner that is an integer linear program, an approval prediction, wherein the approval prediction indicates that the request is to be approved based on one or more launch parameters of the request and the capacity data; receiving information from a downstream component that controls the resource that the approval prediction is incorrect; and evaluating the incorrect approval prediction using a hindsight learner and predictor explainer.