A novel deep reinforcement learning scheme for task scheduling in cloud computing

Recently, the demand of cloud computing systems has increased drastically due to their significant use in various real-time online and offline applications. Moreover, it is widely being adopted from research, academia and industrial field as a main solution for computation and storage platform. Due to increased workload and big-data, the cloud servers receive huge amount of data storage and computation request which need to be processed through cloud modules by mapping the tasks to available virtual machines. The cloud computing models consume huge amount of energy and resources to complete these tasks. Thus, the energy aware and efficient task scheduling approach need to be developed to mitigate these issues. Several techniques have been introduced for task scheduling, where most of the techniques are based on the heuristic algorithms, where the scheduling problem is considered as NP-hard problem and obtain near optimal solution. But handling the different size of tasks and achieving near optimal solution for varied number of VMs according to the task configuration remains a challenging task. To overcome these issues, we present a machine learning based technique and adopted deep reinforcement learning approach. In the proposed approach, we present a novel policy to maximize the reward for task scheduling actions. An extensive comparative analysis is also presented, which shows that the proposed approach achieves better performance, when compared with existing techniques in terms of makespan, throughput, resource utilization and energy consumption.