Message Queue (MQ) services play a vital role in modern distributed systems as they enable asynchronous communication between services and facilitate the decoupling of various components of the system. Among the many MQ services available, Kafka, Apache Pulsar, Artemis, and RocketMQ are popular choices, each offering unique features and capabilities. As the adoption of MQ services continues to grow, choosing the appropriate service that can meet the requirements of the system has become increasingly challenging. Therefore, a comprehensive comparison of these services is crucial to determine the most suitable one for a specific use-case. This research paper presents a thorough evaluation of these MQ services based on critical metrics such as CPU utilization, memory usage, garbage collection, latency, and throughput. Based on our extensive review, no other research has delved into such a detailed evaluation, thereby establishing our work as a cornerstone in this field. The results of our study offer valuable insights into the strengths and limitations of each service. Our findings indicate that each message queue behaves differently inside the Java Virtual Machine (JVM). This work aims to assist developers and researchers in strategically deploying and optimizing MQ services based on specific system and use-case requirements. In addition to providing machine metrics, our results demonstrate the performance of each message queue under different load scenarios, making it a valuable resource for those seeking to ensure the effective functioning of their MQ services.
Transitioning from monolithic systems to cloud-native based on microservice architecture is essential for organizations facing dynamic technological shifts and growing scalability demands. This paper explores a machine learning-driven approach to decompose monolithic systems into microservices, targeting maintainability and modularization. Utilizing static analysis, we extract critical dependency data from the monolith, which guides the configuration of a Variational Autoencoder (VAE) and fuzzy c-means clustering process. This approach enables precise tuning of hyperparameters to optimize the decomposition into highly independent, scalable microservices. Our findings highlight the effectiveness of integrating static analysis with machine learning to enhance the adaptability and efficiency of distributed systems, providing valuable insights into the nuanced impacts of hyperparameter adjustments on system performance. Furthermore, we provide a novel system multi-variant benchmark to the community.