Evaluating Nested and Non-Nested Software Transactional Memory Using Machine Learning Classifiers

Abstract

Software Transactional Memory (STM) provides a robust solution for addressing concurrency challenges in software systems. This paper explores the performance evaluation of nested and non-nested STM configurations using a machine learning-based framework. The dataset used for model training was generated through detailed heap profiling of nested and non-nested STM configurations, capturing memory allocation and usage patterns as key indicators of STM behaviour. The framework leverages profiling datasets to analyse STM operations through four machine learning models: Naive Bayes, Decision Tree, K-Nearest Neighbours (KNN), and Random Forest. The methodology includes data preprocessing, model training, and visualization using MATLAB R2020b, with a focus on 20 profiling metrics that encapsulate key STM operations. Computational experiments reveal that Naive Bayes, KNN, and Random Forest achieved 100% accuracy, precision, recall, and F1-score, while Decision Tree showed lower performance. These results demonstrate the potential of machine learning to evaluate STM behaviour through data-driven analysis.