Runtime Microservice Self-distribution for Fine-grain Resource Allocation

The development of systems using microservices as buildingblocks have gained! major popularity in the industry in the past few years. Widely used services, such as Netflix and Uber, have been built entirely as microservice architectures. Due to the modularity and self-containedness of microservices, coupled with the use of elastic deployment infrastructures, a number of tools to assist the scalability of such systems have been created. However, these tools are limited to act at a fixed granularity, being able to replicate, relocate and provide access to extra resources only at the level of the entire microservice, even when only one of its parts actually demands more resources. In this paper, we propose the use of the concepts of adaptive component models, emergent microservices, and self-distributing systems to explicitly define the internal micro-architecture of microservices. In this approach, a microservice is built as a dynamic configuration of components, which can be seamlessly adapted and distributed on top of an elastic cloud infrastructure by the underlying platform. We evaluate the benefits of the approach by exploring different scenarios that entail the use of dynamic adaptation and self-distribution to perform vertical and horizontal scaling of microservices at a fine granularity. We analyze the involved tradeoffs and discuss how the approach can be further explored.