Bottleneck Detection And Performance Prediction For Large-scale Complex Systems

As the complexity of large-scale enterprise applications increases, multi-tier distributed applications play an increasingly crucial role in business-critical enterprise operations. Understanding the performance of such applications is challenging yet essential due to their growing economic value. In this paper, based on TPC-App benchmark, we discuss and research the mothed of bottleneck detection and performance prediction for large-scale complex systems.We use decision tree classification ID3 algorithm, and propose a machine learning-based automated bottleneck detection method for multi-tier distributed systems. In order to verify the accuracy of this method, we implement TPC-App benchmark to simulate the activities of a business-to-business transactional application server operating. Experimental results show that the method has strong applicability and versatility.In order to effectively and accurately answer the following question: How will performance change in response to variations of workloads in multi-tier applications? We propose a novel statistical regression-based approach to predict performance. And for the purpose of investigating factors which influence the accuracy of the proposed performance prediction method, we change the number of application server nodes (from 1 to 9) and use different degree (from 1 to 5) of regression equation. Moreover, we apply our prediction method to the circumstances where only a small amount of training data is provided to validate its practicability and efficiency. Our empirical results show that this statistical regression-based method provides a practical and powerful solution that can further be used for performance prediction and capacity planning of large-scale complex systems.