Improving an Artificial Hormone System’s Time Bounds Using Task Allocation Signals

Abstract

This paper proposes an extension to the Artificial Hormone System (AHS) in order to improve its time bounds. The AHS is a middleware to distribute tasks to a distributed system’s nodes in a decentralized manner based on Organic Computing principles. The AHS’ task distribution has previously been extended by a priority-based approach that allows to prioritize tasks, allowing to control the order of their allocation during the system’s initial self-configuration or self-healing after a node failure. However, this extension worsens the hard time bounds guaranteed by the AHS compared to its original implementation without priority support. We thus propose so-called task allocation signals that improve these time bounds. This approach results in a speedup factor of up to (2 − 1/m) where m is the number of tasks to distribute in the system, restoring the original AHS’ time bounds while keeping the priority support intact. Evaluations conducted using a cycle-accurate AHS simulator fully confirm the theoretical results presented.