Fault tolerance in distributed systems using deep learning approaches.

Area of Science:

• Computer Science
• Artificial Intelligence
• Distributed Systems

Background:

• Distributed systems are foundational for modern technologies like blockchain and IoT.
• Fault tolerance and decentralization are key features of distributed systems.
• Deep learning excels at pattern recognition for data analysis tasks.

Purpose of the Study:

• To investigate the application of deep learning for fault detection and correction in distributed systems.
• To evaluate deep learning model performance across three distinct fault scenarios.
• To analyze the impact of faulty data size on model accuracy for structured and unstructured datasets.

Main Methods:

• Employed deep learning models including VGG16, VGG19, AlexNet, LSTM, and ResNet34.
• Tested models on three fault scenarios: faulty output, corrupted inputs, and unrelated data patterns.
• Evaluated performance using both structured and unstructured datasets with varying proportions of faulty data.

Main Results:

• Deep learning models successfully identified and corrected faults in distributed systems across all tested scenarios.
• Model accuracy for unstructured datasets ranged from 60% to 96% depending on faulty data size.
• Structured datasets showed high resilience, with accuracy reaching 99% regardless of faulty data portion.

Conclusions:

• Deep learning offers a robust solution for managing faults in distributed systems.
• Model performance is sensitive to the volume of faulty data in unstructured datasets.
• Deep learning effectively handles diverse fault types, including novel patterns, in distributed environments.