Hållbarhet i webbarkitektur: En komparativ studie av energiförbrukning i monolitisk arkitektur kontra microservicearkitektur.

Sammanfattning

The rapid growth of information and communication technology has established energy efficiency as a central architectural quality attribute. The choice of software architecture directly affects how systems utilize underlying hardware resources, but there is a lack of systematic mapping of energy consumption per request across a continuous load spectrum. This study investigates how monolithic architecture and microservice architecture differ from an energy consumption perspective at low, medium and high load levels. The study also investigates whether there is a load level where microservices achieve equivalent or lower energy consumption than the monolithic architecture. A quantitative method with an experimental research design is applied where Spring Petclinic is used as the system artifact in both architectural versions. Both artifacts are run using Docker containers in an isolated test environment on an Intel-based system running Ubuntu. Energy consumption is measured via Intel RAPL using Scaphandre, and load generation is conducted with Locust. Each load level is executed for 30 iterations, and the results are standardized to joules per request. The results indicate that the microservice architecture consumed more energy per request than the monolithic architecture at all tested load levels. At low and medium loads, the difference amounted to approximately 11 percent, and at high load it increased to approximately 12 percent. No break-even point could be identified within the tested spectrum. The trend rather points toward a widening gap as load increases. Beyond the average energy values, the microservice architecture exhibited a progressively increasing variability. The standard deviation ratios relative to the monolithic architecture were 2.50 at low load, 2.71 at medium load and 4.09 at high load. The increasing overhead is explained by network communication, serialization, deserialization, and the operation of multiple JVM instances. The results should not be interpreted as a general argument against microservice architecture, whose advantages in terms of scalability, fault isolation and independent deployment are well-founded. The study instead provides an empirical basis for the energy dimension that should be considered when making architectural decisions, particularly for systems deployed without active horizontal scaling.