Edge Fault Model Selection for Resource-Constrained Cloud Sites
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional Fault Management Systems (FMS) in edge cloud systems require manual reconfiguration when infrastructure changes, which is inefficient for dynamic environments, and deploying multiple Machine Learning (ML) models consumes excessive resources, making it impractical for resource-constrained edge sites.
Innovation Solution
A method and apparatus for selecting Fault Management Models (FMMs) at edge sites based on probability functions of fault occurrence, using a Deployment Agent (DA) to determine the number and type of FMMs to run, inspired by antlion behaviors, to balance fault handling and resource utilization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple ML models are deployed to handle all possible faults, then fault detection and prediction capability is improved, but resource consumption (computing, storage, network) increases significantly
Solution Approach 1:
The patent implements dynamic model selection where the FMS adapts its behavior based on real-time system state. The system transitions between different operational modes (threshold-based, ML-based, hybrid) depending on fault probability assessments, resource availability, and system conditions. This dynamic adaptation allows the system to optimize the balance between fault detection capability and resource consumption by activating ML models only when necessary rather than running them continuously.
Solution Approach 2:
The system changes operational parameters dynamically by adjusting which fault management approach is active based on assessed fault probability. When fault probability is low, the system switches to lower-resource threshold-based monitoring. When probability increases, it transitions to ML-based detection. This parameter change enables the system to maintain high reliability when needed while minimizing resource consumption during normal operation.
2Speed
If ML models are run online for timely inferencing, then fault prediction speed is improved, but computational overhead increases compared to traditional threshold-based methods
Solution Approach 1:
The system employs periodic assessment of fault probability and system state to determine when to activate ML models. Rather than continuous ML inference, the system periodically evaluates whether conditions warrant switching to ML-based detection, thereby reducing computational overhead while maintaining the ability to provide timely predictions when faults are likely.
Solution Approach 2:
The patent introduces a hybrid architecture where threshold-based monitoring serves as an intermediary layer that filters normal operations. This intermediary continuously monitors system parameters and only triggers computationally intensive ML models when threshold violations or suspicious patterns are detected, thereby reducing overall computational overhead while maintaining fast response capability.
3Stability of the object's composition
If conventional FMS is deployed with manual reconfiguration, then system stability is improved, but adaptability to infrastructure changes deteriorates
Solution Approach 1:
The system implements self-service through automated model selection and configuration. The FMS autonomously assesses system state, fault probability, and resource availability to determine which models to deploy and how to configure them. This self-service capability enables the system to adapt to infrastructure changes automatically without manual reconfiguration, while the gradual adaptation process maintains system stability.
Solution Approach 2:
The patent incorporates feedback mechanisms where the system continuously monitors performance metrics, fault detection accuracy, and resource consumption. This feedback loop allows the FMS to learn from past performance and automatically adjust its model selection and configuration strategies, enabling adaptation to infrastructure changes while maintaining stability through data-driven decision making.
Data Source
AI summary
Embodiments described herein provide methods and apparatuses for selecting Nnext Fault Management Models, FMMs, where Nnext is an integer value, to run at an edge site in a cloud system during an ith time period. A method in a Fault Management System, FMS includes: in a deployment agent, DA: selecting the Nnext FMMs from a first set of FMMs based at least in part on respective probability functions associated with each of the first set of FMMs, wherein each respective probability function indicates how likely a fault associated with the respective FMM is to occur as a function of time; and initiating running of the Nnext FMMs at the edge site during the ith time period.