Adaptive online task scheduling algorithm for resource regulation on heterogeneous platforms

As computing technology advances, resource regulation on heterogeneous platforms has emerged as a key research area for future computing environments. In cloud task scheduling, studies focus on intelligent agent models and performance indicators that balance user experience and cost-effectiveness. Research into deep reinforcement learning and deep deterministic policy gradient (DDPG) algorithms has been conducted, incorporating heterogeneous resource regulation to address the varied needs of different data centres. Key task characteristics include length, average instruction length, and average CPU utilisation, with significant standard deviations. During training, a Poisson distribution parameter with a lambda value of 1 was used, leading to convergence in the loss curve. Although the DDPG algorithm had a slightly higher virtual machine usage cost and an instruction response time of 306.5, it provided notable economic benefits, demonstrating improved management and utilisation of computing resources.