IntelliScheduler: an edge-cloud computing environment hybrid deep learning framework for task scheduling based on learning

Overview

IntelliScheduler is a hybrid deep reinforcement learning framework designed for adaptive task scheduling in edge-cloud computing environments. The system addresses the challenge of maintaining Service Level Agreement (SLA) compliance and quality of service (QoS) in heterogeneous IoT workflow applications characterized by varying deadlines and dynamic workloads. The framework integrates actor-critic deep reinforcement learning with runtime-aware state representation and multi-buffer experience replay architecture to enable adaptive decision-making across distributed computational resources.

Methods and approach

The research develops a learning-based optimal task scheduling (LbOTS) algorithm that employs actor-critic deep reinforcement learning to minimize total task execution delay. The framework incorporates runtime-aware state representation that captures system conditions and a decision mechanism informed by latency-aware reward modelling. The architecture utilizes multi-buffer experience replay to improve learning stability and convergence. Task deployment decisions are optimized across heterogeneous edge and cloud resources through the learned policy, with the objective of discovering optimal placement strategies that minimize end-to-end latency while maintaining resource efficiency.

Key Findings

Experimental evaluation conducted through simulation demonstrates superior performance compared to particle swarm optimization (PSO), multi-objective particle swarm optimization (MOPSO), and mating behavior optimization (MBO) baselines. The LbOTS algorithm achieves approximately 13% higher normalized reward, 67% lower training loss, 52-66% lower operational cost, and 80-90% lower task rejection rate. Quality of experience (QoE) improvements range from approximately 15-75% relative to baseline approaches across various experimental scenarios and workload configurations.

Implications

The adaptive learning formulation presented in IntelliScheduler addresses fundamental limitations of cloud-centric and heuristic scheduling approaches by enabling dynamic responsiveness to fluctuating system conditions. The demonstrated improvements in SLA compliance, operational efficiency, and task rejection rates indicate that reinforcement learning-based methods can effectively handle the complexity and variability inherent in edge-cloud computing environments. The framework's capacity for runtime adaptation suggests potential for deployment in production systems where workload characteristics and resource availability change continuously.

Scope and limitations

This summary is based on the study abstract and available metadata. It does not include a full analysis of the complete paper, supplementary materials, or underlying datasets unless explicitly stated. Findings should be interpreted in the context of the original publication.