Predicting Effect and Cost of Microservice System Evolution Using Graph Neural Network

Abstract

With the increasing prevalence of microservice technology, the architectural flexibility and scalability of software systems have witnessed notable advancements. However, this progress has also brought about a challenge in meeting the frequent changes in user requirements, thereby adversely affecting the quality of the system. It is crucial for microservice systems to undergo evolution through the modification of system configurations to adapt to changing requirements, and various methods for system evolution have been proposed. However, the evolution schemes generated by these methods vary in terms of the degree of improvement in quality and the cost required for evolution, such as time and money, i.e., different evolution effect and evolution cost. Considering the above, it is necessary to predict effect and cost before applying these schemes to real systems. Existing physical methods possess drawbacks such as high expenses and time-consuming setup procedures. Conversely, simulation methods, which are based on mathematical models, necessitate certain simplifications, resulting in disparities between the outcomes and the actual results. To overcome these challenges, this paper introduces a prediction method for microservice system evolution. By employing Graph Neural Network techniques to learn from historical data, this method enables precise prediction of the effects and costs associated with various microservice evolution schemes. And based on the above algorithm, an online prediction system is implemented, independent of the microservice system for long-term prediction. Experimental results validate the accuracy and robustness of the proposed prediction method.