Distributed systems can consist of thousands of network nodes interacting with each other. Given their size, managing these systems to perform optimally is a task that should be automated. Multi-agent reinforcement learning (MARL) is a suitable technique for tackling such problems. However, the application of MARL in a distributed system is not trivial. To bridge the gap between these two domains, we introduce the Marlon language. It enables MARL experts to focus on solving machine learning problems, rather than the complexities of distributed computing. We evaluate Marlon by comparing the implementation of a load balancing use case in Marlon with an ad-hoc implementation.
Original languageEnglish
Title of host publication34th ACM/SIGAPP Symposium On Applied Computing
Number of pages8
Publication statusPublished - 2019

ID: 40531532