Markov Chain-Based Resource-Efficient and QoS-Aware Scheduling for Latency-Critical and Best-Effort Tasks

ABSTRACT

Cloud and edge computing environments are increasingly required to support heterogeneous workloads, where latency-critical applications coexist with best-effort tasks under limited computational and communication resources. Ensuring Quality of Service (QoS) while maintaining high resource efficiency has become a significant challenge, particularly in dynamic and stochastic system conditions.

This work proposes a Markov Chain-based scheduling framework designed to intelligently allocate resources between latency-critical and best-effort tasks. The system models task states and transitions using discrete-time Markov chains, enabling probabilistic prediction of workload dynamics and system congestion levels. By leveraging state transition probabilities, the scheduler dynamically prioritizes latency-sensitive tasks while opportunistically utilizing residual resources for best-effort processing.

The proposed approach incorporates QoS-awareness by embedding delay constraints and service requirements directly into the state modeling and decision-making process. A reward-based mechanism is integrated to balance throughput maximization with latency minimization, ensuring that critical tasks meet stringent deadlines without significantly degrading overall system performance.

Simulation results demonstrate that the proposed scheduler achieves improved latency guarantees for critical tasks, enhanced resource utilization, and better fairness compared to conventional scheduling techniques. The framework is particularly suitable for real-time applications such as industrial automation, autonomous systems, and next-generation communication networks where both efficiency and reliability are essential.

Keywords: Markov chain model, resource contention, latency-critical (LC) task, best-effort (BE) task, performance profiling, resource allocation optimization, resource contention scheduling.