供应
计算机科学
云计算
分布式计算
工作流程
强化学习
调度(生产过程)
捆绑
人工智能
数据库
计算机网络
操作系统
运营管理
复合材料
经济
材料科学
作者
Jian Cao,Jian Cao,Rajkumar Buyya
标识
DOI:10.1109/tc.2022.3191733
摘要
The efficient cloud resource provisioning for the execution of complex workflow applications has always been one of the important research issues. Most of the existing approaches focus on the resource provisioning of single-type virtual machine (VM) instances for the single or multiple workflows, while few consider the situation of provisioning multi-type VM instances simultaneously. As a result, the executing performance of complex workflows degrades. Different from the existing work, this paper proposes an adaptive cloud bundle provisioning and multi-workflow scheduling model to dynamically perform both the horizontal and vertical cloud resource scaling on multi-type VM instances for the execution of complex workflows. Among the model, a depth-first-search coalition reinforcement learning (DFSCRL) provisioning policy is presented to realize the resource scaling, which integrates the physical machine (PM) coalition formation with the Q-learning algorithm, then dynamically generates an optimal multi-type VM instance bundle from the PM coalition, and finally provisions these instances to the concurrent execution of multiple workflows. The theoretical proofs and various experiments with the multifaceted metrics demonstrate that the performance of the proposed algorithms is superior to that of the state-of-the-art relevant policies.
科研通智能强力驱动
Strongly Powered by AbleSci AI