Improving software performance and reliability in a distributed and concurrent environment with an architecture-based self-adaptive framework

More and more, modern software systems in a distributed and parallel environment are becoming highly complex and difficult to manage. A self-adaptive approach that integrates monitoring, analyzing, and actuation functionalities has the potential to accommodate an ever dynamically changing environment. This paper proposes an architecture-level self-adaptive framework with the aim of improving performance and reliability. To meet such a goal, this paper presents a Self-Adaptive Framework for Concurrency Architectures (SAFCA) that consists of multiple well-documented architectural patterns in addition to monitoring and adaptive capabilities. With this framework, a system using an architectural alternative can activate another alternative at runtime to cope with increasing demands or to recover from failure. Five adaptation mechanisms have been developed for concept demonstration and evaluation; four focus on performance improvement and one deals with failover and reliability enhancement. We have performed a number of experiments with this framework. The experimental results demonstrate that the proposed adaptive framework can mitigate the over-provisioning method commonly used in practice. As a result, resource usage becomes more efficient for most normal conditions, while the system is still able to effectively handle bursty or growing demands using an adaptive mechanism. The performance of SAFCA is also better than systems using only standalone architectural alternatives without an adaptation scheme. Moreover, the experimental results show that a fast recovery can be realized in the case of failure by conducting an architecture switchover to maintain the desired service.