Rapid Prototype for Shifting HPC to the Cloud
Rapid Prototype for Shifting HPC to the Cloud
复制标题
将 HPC 转移到云端的快速原型
DOI：
--
复制
发表时间：
2021
期刊：
影响因子：
--
通讯作者：
Alex Younts
中科院分区：
文献类型：
--
作者：
Sam Weekly;Zoey Mertes;Alex Younts
随着云计算的日益普及，我们已经看到快速变化的环境正在寻找独特的方式来利用内部部署的批调度HPC环境中的云资源。普渡大学的研究计算在为校园内的研究人员提供低成本的高性能计算和存储资源方面有着悠久的历史。我们最近针对云采用的两个领域是大规模工作负载爆发和快速资源原型。利用微软Azure中新的HBv3实例类型，我们能够更快地访问最新的第三代AMD EPYC cpu，而不是交付和部署我们等效的现场资源，即将到来的NSF资助的Anvil超级计算机。这种快速访问使我们的同事和系统设计人员能够完成各种任务，以评估和验证新一代硬件的预期性能，这些硬件将与现场部署并行使用。
With the growing adoption of cloud computing, we have watched the rapidly changing landscape looking for unique ways to leverage cloud resources in an on-premise batch scheduled HPC environment. Research Computing at Purdue University has a long history of providing low cost HPC compute and storage resources to researchers on campus. Two areas we have recently targeted for cloud adoption are large scale workload bursting and rapid resource prototyping. Utilizing the new HBv3 instance types in Microsoft Azure, we were able to access the most recent third generation of AMD EPYC CPUs sooner than the delivery and deployment of our equivalent on-site resource, the upcoming NSF funded Anvil supercomputer. This rapid access allowed our colleagues and system designers to complete various tasks needed to evaluate and verify the expected performance of the newer generation hardware that will be available with Anvil in parallel to the on-site deployment.