Virtualization of stateful services via machine learning

Today’s enterprise software systems are much more complicated than the past. Increasing numbers of dependent applications, heterogeneous technologies, and wide usage of Service-Oriented Architectures (SOA), where numerous services communicate with each other, makes testing of such systems challenging. For testing these software systems, the concept of service virtualization is gaining popularity. Service virtualization is an automated technique to mimic the behavior of a given real service. Services can be classified as stateless or stateful services. Many services are stateful in nature, yet virtualization of stateful services is harder than virtualization of stateless services. In this work, we introduce two novel stateful service virtualization approaches. We employ classification-based and sequence-to-sequence-based machine learning algorithms in developing our solutions. Classification is a supervised learning method where the task is assigning given inputs to corresponding classes. A sequence-to-sequence model is a deep neural network architecture where the input and the output are sequences. We demonstrate the validity of our approaches on three datasets. Our evaluation shows that we obtain 75 % to 81 % accuracy on subject datasets with classification based method. Our deep neural network-based solution achieves even better accuracy results ranging from 89 to 97 % on subject datasets. Our evaluation on training times of the mentioned techniques show that classification based technique significantly outperforms other methods.