Composite Risk Scoring for Node-Level Cloud Deployments

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing risk assessment methods for cloud-based software deployments overlook node-level nuances, leading to inaccurate risk calculations and potential deployment failures, especially when deploying significant software components like host operating systems, due to reliance on cluster-level assessments that fail to account for individual node diversity and specific requirements.

Innovation Solution

A risk assessment framework that utilizes node-level feature data to generate a training dataset, employing machine learning models like XGBoost and LGBM, to calculate deployment risk scores, including constituent risks for interruptions and malfunctions, and accounts for virtual machine size and hardware configurations, enhancing accuracy by capturing individual node nuances.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If cluster-level risk assessment is used, then the assessment process is simplified, but the accuracy of risk calculation deteriorates due to overlooking node-level nuances

Engineering Contradiction:
Improveassessment process complexityVSAvoidrisk calculation accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent segments the risk assessment process into two distinct levels: cluster-level assessment and node-level assessment. The system first performs a preliminary risk assessment at the cluster level to identify potentially affected clusters, then conducts detailed node-level assessments only for those specific clusters. This segmentation allows the system to maintain assessment accuracy by examining individual node characteristics while avoiding the inefficiency of assessing every node in every cluster uniformly.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by tailoring the depth and scope of risk assessment to specific locations (clusters and nodes) based on their individual characteristics. The system evaluates node-level features such as hardware configurations, software versions, and operational status only for nodes within clusters identified as potentially affected. This localized approach ensures that risk calculations are as accurate as possible for each specific node while avoiding unnecessary assessment overhead in low-risk areas.

Inventive Principle:
Principle #3Local quality

2Measurement precision

If node-level feature data is collected and analyzed, then the accuracy of risk assessment improves, but the computational complexity and data processing requirements increase

Engineering Contradiction:
Improverisk assessment accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary risk assessment at the cluster level before conducting detailed node-level analysis. This preliminary action identifies which clusters are potentially affected by the software deployment, allowing the system to focus computational resources only on relevant clusters and their constituent nodes. By filtering out low-risk clusters beforehand, the system significantly reduces the overall computational burden while maintaining comprehensive risk assessment accuracy for high-risk areas.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements a two-tiered assessment approach where full node-level analysis is performed only on clusters identified as potentially affected, rather than conducting exhaustive analysis on all clusters. This partial action approach ensures that the necessary level of detail is applied where needed to achieve accurate risk assessment, while avoiding the excessive computational overhead of uniform comprehensive analysis across the entire infrastructure.

Inventive Principle:
Principle #16Partial or excessive action

3Reliability

If comprehensive risk factors are evaluated, then the reliability of deployment decisions improves, but the time required for risk assessment increases

Engineering Contradiction:
Improvedeployment decision reliabilityVSAvoidrisk assessment time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent segments the risk assessment process into sequential phases: cluster-level screening followed by node-level detailed assessment. This segmentation enables the system to quickly eliminate low-risk clusters through preliminary evaluation, then concentrate time and resources on comprehensive analysis of only those clusters with potential risks. As a result, deployment decisions maintain high reliability for affected clusters while the overall assessment time is significantly reduced compared to uniform comprehensive analysis.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary cluster-level risk assessment as a filtering step before detailed node-level evaluation. This preliminary action quickly identifies potentially affected clusters using broader risk factors, allowing the system to then apply more time-consuming comprehensive analysis only where necessary. This approach ensures that deployment decisions for high-risk clusters are based on thorough evaluation of all risk factors, while the total assessment time is managed efficiently through the staged process.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4654017A1Composite risk score for cloud software deployments
Publication Date: 2025.11.26 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP4654017A1 patent drawingFigure 1
  • EP4654017A1 patent drawingFigure 2
  • EP4654017A1 patent drawingFigure 3

AI summary

The techniques described herein provide a risk assessment framework that enhances the functionality of software deployment systems in cloud-based platforms. Generally described, the present techniques evaluate and consolidate various risk scores to classify a given computing cluster within a software deployment strategy. In various examples, a deployment system collects node-level feature data from the computing cluster to generate a dataset to train a prediction model to calculate constituent risk scores. In another aspect, the deployment system aggregates constituent risk scores to determine an overall risk of software failure. Likewise, the deployment system considers diverse criteria such as virtual machine size and virtual machine density to determine an overall impact of software deployment failure. The deployment system then calculates a composite risk score for the computing cluster as a function of the risk of software deployment failure and the impact of software deployment failure.