Process Classification Model for Unused Process Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for identifying unused processes in computing systems are either underinclusive or overinclusive, leading to sub-optimal results in cleaning up defunct (zombie) and unintentionally orphaned processes, which can negatively affect system performance by occupying system resources and limiting available slots in the process table.
Innovation Solution
A computer-implemented method using a process classification model trained with historic transactional data to generate probability scores for current processes, determining if they are unused, and taking actions based on predefined thresholds, while respecting long-running processes, employing logistic regression and feature selection to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current methods are used to identify unused processes, then some defunct processes can be cleaned up, but the methods are either underinclusive or overinclusive leading to sub-optimal results
Solution Approach 1:
The patent transforms the process identification problem from simple state detection to probabilistic classification by introducing multiple parameters (CPU usage, memory usage, process age, I/O activity) and using logistic regression to calculate probability scores. This allows nuanced differentiation between truly unused processes and active processes, resolving the underinclusive/overinclusive dilemma.
Solution Approach 2:
The patent replaces traditional mechanical rule-based process identification with a machine learning classification model. Instead of using fixed thresholds or simple state checks, the system employs logistic regression with feature selection to dynamically assess process status, achieving higher precision without sacrificing productivity.
2Measurement precision
If probability-based classification is used to identify unused processes, then identification accuracy improves, but computational complexity increases
Solution Approach 1:
The patent segments the complex classification problem into manageable components: feature selection identifies relevant parameters, logistic regression provides the probabilistic framework, and threshold-based decision rules simplify the final classification. This segmentation makes the system tractable while maintaining high accuracy.
Solution Approach 2:
The system performs self-service by automatically training the classification model on historical process data and continuously refining feature selection. This automated approach reduces the need for manual configuration and expertise, making the sophisticated model practical to deploy and maintain.
Data Source
AI summary
A method, system, and computer program product are configured to: create a process classification model using historic transactional data of historic processes in a computing system, wherein the process classification model generates a probability that a process is in an unused state; detect current processes in the computing system; generate a probability score of a respective one of the current processes using current transactional data of the respective one of the current processes with the process classification model; and perform an action with the respective one of the current processes based on the probability score of the respective one of the current processes exceeding a predefined threshold.