Automated Event Generation for Faster ML Failure Prediction Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI/ML-based troubleshooting solutions for data storage failures require extensive training times and struggle to adapt to environmental changes, leading to delayed issue resolution and false positives.
Innovation Solution
A supervised learning approach using automated event generation and data labeling (AEGDL) to generate a labeled dataset through test cases in test configuration clusters, incorporating telemetry information from compute, network, and storage devices to train prediction models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing AI/ML-based troubleshooting solutions are used, then model training can be performed, but training time is extensive and the models are inflexible
Solution Approach 1:
The system performs preliminary actions by automatically generating synthetic failure events and labeling them with ground truth information before actual troubleshooting is needed. This pre-generation of training data with known labels eliminates the need for extensive manual data collection and labeling, thereby reducing training time while maintaining model accuracy.
Solution Approach 2:
The system enables self-service by using automated event generation and labeling mechanisms that do not require human intervention. The AEGDL manager automatically creates synthetic events, executes test cases, collects telemetry data, and generates labeled training datasets, making the entire data preparation process autonomous and significantly faster than manual methods.
2Adaptability or versatility
If existing troubleshooting models are deployed, then predictions can be made, but the models generate false positives and fail to adapt to environmental changes
Solution Approach 1:
The system implements dynamics by continuously generating new synthetic failure events that reflect changing environmental conditions and configurations. The automated event generation creates diverse scenarios including hardware failures, software bugs, and configuration errors across different device types and configurations, enabling models to adapt to environmental changes while maintaining high prediction accuracy through continuous exposure to varied training examples.
3Quantity of substance
If manual data collection and labeling is performed, then training datasets can be created, but the process is time-consuming and labor-intensive
Solution Approach 1:
The system performs self-service by automatically generating synthetic failure events with known ground truth labels, eliminating the need for manual data collection and labeling. The AEGDL manager autonomously creates diverse failure scenarios, executes test cases, collects telemetry data, and generates labeled datasets in high volumes without human intervention, dramatically reducing both time and labor requirements while producing large quantities of training data.
Solution Approach 2:
The system uses copying by creating synthetic replicas of real-world failure scenarios through automated test case execution. Instead of manually collecting and labeling actual failure data, the system generates copies of failure events with known outcomes, preserving the characteristics of real failures while eliminating the time-consuming manual labeling process and enabling rapid dataset generation.
Data Source
AI summary
A method for generating testing information for prediction model training includes: identifying, by an automated event generation and data labeling (AEGDL) manager, a test case initiation event; in response to the identification: selecting a first test case of a plurality of test cases; identifying a first test configuration cluster (TCC) of a plurality of TCCs based on the test case; performing the first test case in the TCC; generating testing information based on the performing of the first test case in the TCC; and providing the testing information to prediction models for model training.