Parallel random matrix particle swarm optimization scheduling algorithms with budget constraints on cloud computing systems
Parallel random matrix particle swarm optimization scheduling algorithms with budget constraints on cloud computing systems
复制标题
DOI：
10.1016/j.asoc.2021.107914
复制
发表时间：
2021-12
期刊：
影响因子：
--
通讯作者：
Xiaoyong Tang;Cheng Shi;Tan Deng;Zhiqiang Wu;Li Yang
中科院分区：
文献类型：
--
作者：
Xiaoyong Tang;Cheng Shi;Tan Deng;Zhiqiang Wu;Li Yang
如今，越来越多的物联网和移动互联网应用服务迁移到云计算系统。该业务最重要的云挑战之一是优化服务成本。应对这一挑战的有效方法是提高云系统中资源管理和任务调度的性能。然而，这种服务调度是典型的组合优化问题，很难得到最优解。在本研究中，我们首先在具有预算约束调度问题的虚拟机（VM）上形式化这种云服务。然后，提出一种随机矩阵粒子群优化调度算法（RMPSO），该算法使用随机整数矩阵来表示其位置和可行的任务调度方案，以实现云服务总成本的最优。然而，这种解决方案对于大规模系统的缺点是时间复杂度较高。因此，我们提出了两种并行RMPSO算法：具有共享内存的多核系统上的CPU并行算法（M-RMPSO）和众核GPU加速策略（G-RMPSO）以降低其时间复杂度。最后，严格的性能评估结果清楚地表明，我们提出的 G-RMPSO 在云服务总成本和算法执行时间方面优于 M-RMPSO 和现有的 FMPSO、HYBRID（MPSO+MCSO）。因此，我们提出的GPU加速G-RMPSO算法非常适合云服务调度。
Nowadays, increasing number of Internet of Things and mobile Internet application services are migrated to cloud computing systems. One of the most important cloud challenges for this business is to optimize services cost. The efficient way to deal with this challenge is to improve the performance of resource management and task scheduling in cloud systems. However, this kind of service scheduling is a typical combinatorial optimization problem, and it is difficult to obtain the optimal solution. In this study, we first formalize this cloud services on virtual machines (VMs) with budget constraints scheduling problem. Then, we propose a random matrix particle swarm optimization scheduling algorithm (RMPSO), which uses the random integer matrix to represent its position and a feasible task scheduling scheme, to achieve the optimal total cost of cloud services. However, the drawback of this solution for large-scale systems is its high time complexity. Therefore, we propose two parallel RMPSO algorithms: CPU parallel algorithm (M-RMPSO) on multi-core system with shared memory and manycore GPU-accelerated strategy (G-RMPSO) to reduce its time complexity. Finally, the rigorous performance evaluation results clearly show that our proposed G-RMPSO outperforms M-RMPSO and existing FMPSO, HYBRID (MPSO+MCSO) in terms of cloud services total cost and algorithm execution time. Therefore, our proposed GPU-accelerated G-RMPSO algorithm is very suitable for cloud service scheduling.