A Scalable Framework for High-Performance Computing with Cloud
A Scalable Framework for High-Performance Computing with Cloud
复制标题
用于云高性能计算的可扩展框架
DOI：
10.1007/978-981-16-5987-4_24
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
D. Rajeswara Rao
中科院分区：
文献类型：
--
作者：
M. Abhishek;D. Rajeswara Rao
在云环境中，根据用户需求，根据为硬件计算资源提供的配置来配置虚拟机。为了配置这些虚拟机，云环境拥有自己的多台服务器的分布式基础架构。在大小规模上，基础设施即服务正被用作执行高性能计算(HPC)应用的可行平台。然而，高性能计算应用的未来性特征和当前基于云调度的算法仍然存在很少的滞后。本文提出了一个可扩展的框架，它使用基于算法的方法来预测匹配的底层计算资源，以适合运行高性能计算(HPC)应用程序。它将发挥决策者的作用，预测集装箱的位置。它的开发是为了促进云计算环境，为HPC从专用分配的容器中使用适当的计算资源提供解决方案，并使用Kubernetes进行监控。讨论了该框架以及所提出的容器上的解决方案，并用计算算法进行了说明，并给出了结果。它将根据需要扩展或缩减资源，并在几分钟内产生或停用HPC群集。我们使用Kubernetes来监控部署的容器，我们的结果表明共享资源对性能没有影响。
In cloud environment, virtual machine (VM) is provisioned depending on the configuration provided for the hardware computing resources based on user requirement. In order to provision these VMs, cloud environment has its own distributed infrastructure of several servers. On major and minor scale, infrastructure as a service is getting used as a feasible platform for the execution of high-performance computing (HPC) applications. However, few lags still exist within the futuristic traits of HPC application and current cloud scheduling-based algorithms. This paper presents a scalable framework that uses an algorithm-based approach to predict the matching underlying computing resource suitable to run high-performance computing (HPC) applications. It will behave like a decision-maker to predict the placement of containers. It is developed to facilitate a cloud computing environment to provide solution for availing the appropriate computing resources from dedicated allocated containers for HPC and monitoring of the same using Kubernetes. The framework along with proposed solution over containers is discussed and illustrated by computational algorithm with results. It will scale up or down the resources as required and will spawn or decommission the HPC clusters within minutes. We use the Kubernetes to monitor the deployed containers, and our results indicate that shared resources are having no impact on performance.