露天矿山智能调度管理系统可靠性研究

Research on the Reliability of Intelligent Dispatching Management System for Open-pit Mines

可靠性是衡量一个系统能否稳定成功运作的重要指标,高可靠性意味着系统在绝大部分时间都能正常运作,反之亦然。露天矿山智能调度管理系统(简称“调度系统”)通常被部署在云平台上,具有复杂的结构和多样的部署方式,因此难以获得其可靠性信息。为此,文章提出一种基于蒙特卡罗仿真的可靠性评估方法,其将该调度系统类比为一个混合云应用,并将调度系统部署栈建模为考虑相对全面的组件及其依赖关系的分层依赖关系图,通过蒙特卡罗方法对各个组件进行仿真,再根据分层依赖图确定调度系统的状态。同时,考虑到不同服务实例部署方式对调度系统可靠性有着关键影响力,文章提出了一种逐步遍历算法,以获取最佳的服务实例部署方式,进而得到最可靠的调度系统部署方式。实验结果表明,在不同云服务器选择下,文章所提出的服务实例部署方法均优于随机部署方法与均匀部署方法。以公有云调度系统为例进行实验,其经过遍历部署后,在总服务实例为40时,就可以获得99.99%以上的可靠性;而随机部署方式需要总服务实例达到47时,才能达到99.99%以上的可靠性;并且在总资源较少时,遍历部署的调度系统最多可比随机部署的调度系统高出3.87%的可靠性数值。

Reliability is an important indicator of whether a system can operate stably and successfully, high reliability means that the system can operate normally most of the time, and low reliability is the opposite. The intelligent scheduling system for open-pit mines is usually deployed on the cloud platform, and has a complex structure and a variety of deployment methods, so it is difficult to obtain its reliability information. To this end, this paper proposes a reliability evaluation method based on Monte Carlo simulation, which compares the scheduling system to a hybrid cloud application, and models the scheduling system deployment stack as a hierarchical dependency diagram considering relatively comprehensive components and their dependencies, simulates each component through the Monte Carlo method, and then determines the status of the scheduling system according to the hierarchical dependency diagram. At the same time, considering that different service instance deployment methods have a key impact on the reliability of the scheduling system, this paper proposes a step-by-step traversal algorithm to obtain the best service instance deployment method, and then obtain the most reliable scheduling system deployment method. Experimental results show that the service instance deployment methods proposed in this paper are better than the random deployment method and the uniform deployment method under different ECS selections. Taking the public cloud scheduling system as an example, after traversal deployment, when the total service instance is 40, it can obtain more than 99.99% reliability. The random deployment method requires the total service instance to reach 47 to achieve more than 99.99% reliability;and when the total resources are small, the traversal deployed scheduling system can be up to 3.87% higher than the randomly deployed scheduling system.