Evaluation of microservice implementation approaches for image processing

In recent years we can observe an increasing interest in the relatively young topic of Microservice Architectures. As a consequence thereof, the number of experiences with Microservices is increasing and both advantages, as well as disadvantages become a commonly discussed topic. While first best practices and architectural patterns regarding the construction of such systems emerged, there is a lack of discussion regarding the construction and applicability in certain contexts of Microservice Systems. The current state of knowledge revolves primarily around general properties of microservices and therefore use cases in contexts with special characteristics are still uncharted terrain. This thesis provides a closer look at Microservices in the context of image processing systems. The evaluation was performed based on a system offering five commonly used image transformations. With a given set of requirements this system was implemented without the use of a framework, with a framework and with an Function as a Service approach to achieve a Microservice Architecture. All three of the resulting applications were evaluated via quantitative metrics such as the time spent on development and performance characteristics. A qualitative evaluation was performed as well, which compared advantages and disadvantages of the different approaches and yielded findings regarding Microservice Architectures in the context of image processing. The results showed that no single, generally preferable approach exists for such a system. All three implementations offered different advantages and disadvantages in the form of a trade-off between flexibility and development speed. However, there were indications that generally a Function as a Service approach can yield the most benefits, because of a strong reduction in both development time and the complexity of the resulting source code. The combination of image processing and Microservices lead to properties which had a strong influence on the resulting architecture. This was primarily caused by the long processing time of the requests and the high load on the network caused by the transfers of image files between services. As a result thereof, commonly used and simple approaches would often have resulted in serious disadvantages, and more complex mechanisms such as asynchronous communication were necessary throughout the whole system. However, it became apparent that a correct application of microservice principles leads to better maintainability and scalability of such a system.