Title: Overview: A Generative Markov Model Approach for Distributed Computing Infrastructures

Abstract: New distributed computing frameworks, including the computing continuum, are characterized by their intrinsic complexity, stochasticity, and heterogeneity. To maximize the efficiency and effectiveness of resource utilization throughout these expansive systems, there is a critical need for a cohesive formal modeling approach. This paper introduces a comprehensive framework that represents distributed computing systems as a generative Markov model, structured around a defined system state. Within this architecture, the system state is broken down into high-dimensional variables, which are subsequently factorized by their individual components. This decomposition captures the sparse dependency patterns typical of distributed environments, resulting in a computationally tractable model. Consequently, the framework facilitates simulation, inference, and policy learning even when dealing with otherwise unmanageable system states, thereby connecting distributed computing with Markov chain theory and reinforcement learning (RL).

The utility of this framework is illustrated through a case study focused on collaborative AI inference. In this scenario, a central server integrates its own computational power with resources contributed by service users. The findings indicate that as the system scales, centralized scheduling emerges as a performance bottleneck. In contrast, distributing computational tasks across user-owned devices leads to decreased latency and lower resource demands on the server. These results underscore the importance of adaptive decision-making mechanisms in distributed systems and validate the proposed framework’s effectiveness for modeling, simulation, and optimization purposes.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC