Data-driven Understanding of Design Decisions in Pattern-based Microservices Architecture
Andres Diaz-Pace, Catia Trubani and
David Garlan.
In Proceedings of the 19th European Conference on Software Architecture (ECSA), Limassol, Cyprus, 15-19 September 2025. To appear.
Online links:   Plain Text
Abstract
| The adoption of architectural patterns has recently been assessed in relation to their impact on the performance of microservice-based applications. For example, offloading common functionalities of
multiple microservices to a gateway may lead to a system response time
improvement. However, for a given system requirement, e.g., the latency
of services or the utilization of resources, the benefit of choosing an architectural pattern is not guaranteed. Therefore, it becomes important
to collect data about the parameters that contribute to the effective use
of patterns, thus understanding the relationships between design decisions and performance requirements. In this work, we propose a data-driven approach to assess the quantitative impact of design decisions
for a given pattern on the achievement of performance tradeoffs. Our
approach seeks to control the pattern parameters that cause variations,
i.e., sensitivity, in performance tradeoffs. Starting from a dataset including parameters related to three microservices patterns (i.e., Gateway
Offloading, Command and Query Responsibility Segregation, and Anticorruption Layer) and their performance characteristics, we do apply
machine learning techniques (i.e., PRIM and CART) to infer constraints
on the parameter values. This is helpful to understand and reduce the
performance sensitivity of pattern configurations. Our results support
software architects in making informed decisions by providing insights
on the parameters related to the behavior of microservices patterns. |
Keywords: Software Architecture.
|
|
