ColonyOS -- A Meta-Operating System for Distributed Computing Across Heterogeneous Platform

Abstract: This paper presents ColonyOS, an open-source meta-operating system designed to improve integration and utilization of diverse computing platforms, including IoT, edge, cloud, and HPC. Operating as an overlay, ColonyOS can interface with a wide range of computing environments, fostering creation of so-called compute continuums. This makes it possible to develop AI workflows and applications that can operate across platforms. At its core, ColonyOS consists of distributed executors that integrate with various underlying platforms based on a distributed microservice architecture. These executors collectively form a colony, serving as a unified computing unit. To enable secure integration of various platforms, each colony is provisioned with precisely the resources needed, and all communication is confined within the colony governed by a strict zero-trust security protocol. Interaction with ColonyOS is done by submitting functional meta-descriptions of computational tasks, called function specifications. These are sent to a Colonies server, which acts as intermediary between applications and the executors. Upon assignment, an executor interprets the meta-description and translates it into an executable format, e.g. a Kubernetes deployment description, a Slurm script, or a direct function call within the executor. Furthermore, a built-in meta-file system enables data synchronization directives to be included in meta-descriptions, enabling seamless data management across platforms. Ultimately, ColonyOS paves the way for development of hyper-distributed applications and workflows, which can seamlessly operate in a computing continuum. The paper describes design principles and implementation details of ColonyOS.