SLAs-Aware Online Task Scheduling Based on Deep Reinforcement Learning Method in Cloud Environment
Longyu Ran, Xiaoyu Shi, Mingsheng Shang · 2019
As increasingly traditional applications migrate to the cloud, load balancing without sacrificing the performances of cloud-based applications is a highly challenging problem in cloud computing. Existing solutions focusing on online task scheduling have the inefficient issue, due to the highly dynamic nature of cloud workloads and the virtual machines (VMs) with heterogeneous hardware configurations. To address this issue, dependent on the strong perception and decision-making ability of deep reinforcement learning (DRL) in automatic control problems, we propose a deep reinforcement learning based task scheduling method, which can dynamic assign submitted tasks to different VMs. In details, we formulate the task scheduling as a dynamical optimal problem with constraints, and then adopt the deep deterministic policy gradients (DDPG) network to find the optimal task assignment solution with meeting the Service Level Agreements (SLAs) requirements. It makes the optimal scheduling decision only dependent on learning directly from its experience without any prior knowledge. The experimental results using the actual Alibaba cloud workload tracings show that compared with other existing solutions, our proposed task scheduling method can effectively reduce the average response time of tasks with guaranteeing the load balancing among VMs, when facing of dynamical workloads.