A Clustering Approach to Cloud Workflow Scheduling using Reinforcement Learning (RL)

Cloud computing is a computational model in which users receive computational resources according to the law of payment in exchange for the use of resources. Satisfying users' variable requests requires an efficient scheduling mechanism due to limited cloud resources. Due to the variety of user requests, task scheduling and service provision have a special place in cloud computing. In this study, a clustering approach based on reinforcement learning (RL) is proposed, which leads to increased system efficiency by futuristic allocation of tasks to resources due to its ability to adapt to the environment and provide appropriate responses to time-varying requests. This algorithm has three processing steps: ۱. clustering, ۲. scheduling policies, and ۳. resource allocation. Besides, in this problem, task scheduling can be considered a Markov decision process (MDP) that includes elements of state space and a set of actions. This model aims to increase resource productivity and justice in the implementation priority. The results modeled in the WorkflowSim simulator show that it leads to improved scheduling and cost parameters.

Keywords:

Cloud Computing , Clustering , Scheduling , Directed Acyclic Graph (DAG) , Reinforcement Learning (RL).