Multitier Web System Reliability: Identifying Causative Metrics and Analyzing Performance Anomaly Using a Regression...

Area of Science:

• Computer Science
• Software Engineering
• System Performance Analysis

Background:

• Multitier Web systems are increasingly complex, making continuous service availability critical for customer satisfaction.
• Current anomaly detection methods focus on single system performance indicators (SPIs) and separate root cause analysis, increasing diagnostic burden.
• Early detection and root cause identification of performance anomalies are essential for maintaining service quality.

Purpose of the Study:

• To propose an advanced causative metric analysis (ACMA) framework for diagnosing performance anomalies in multitier Web systems.
• To develop a method that integrates anomaly detection with root cause analysis for efficient problem resolution.
• To provide a robust framework that can adapt to changes in system performance indicators.

Main Methods:

• Extracted 191 performance metrics (PMs) related to a target SPI.
• Utilized statistical methods to identify 62 vital PMs influencing the target SPI's variance.
• Implemented a random forest regression model to detect causative metrics (CMs) among vital PMs.

Main Results:

• The ACMA framework successfully identified vital PMs and their influence on the target SPI.
• A random forest regression model was developed to detect causative metrics (CMs) responsible for performance anomalies.
• The proposed model demonstrated adaptability, requiring no structural changes even when the target SPI varied.

Conclusions:

• The ACMA framework effectively detects performance anomalies and their causative metrics in enterprise systems.
• Integrating anomaly detection with root cause analysis through ACMA simplifies performance issue diagnosis.
• The ACMA framework offers a scalable and adaptable solution for monitoring complex multitier Web systems.