Abstract
Self-adaptive systems are designed to modify their architecture or
behavior to uphold high-level objectives despite changes in their
operating environments. A critical aspect of developing such systems
involves creating strategies to handle unexpected events in
the operating environments. While this remains an active area of
research within the autonomic computing and self-adaptive systems
community, one commonly adopted approach is leveraging
machine learning techniques, particularly reinforcement learning,
to address unforeseen challenges. In this paper, we conduct experiments
using the EmergentWeb Server exemplar, a publicly available
self-adaptive web server, to investigate various monitoring metrics
and implement a multi-armed bandit reinforcement learning
approach. This approach enables the system to identify the optimal
web server configuration for maximizing performance under
varying workload patterns and operating conditions, enabling the
system to react to unexpected events that rises from the operating
environment with minimum human interference.