5:15 PM - 6:45 PM
[MGI30-P05] A Case Study of Constructing a Computational Foundation for the Global Atmospheric Model on mdx
Keywords:Numerical Model, Virtual Machine, mdx, High Performace Computer
“mdx” is a virtual machine platform based on VMware, equipped with high-speed CPU/GPU, storage, and network. “mdx” is operated by nine national universities and two national research institutions in Japan, and its users are mainly researchers who need a High Performance Computing (HPC) environment.
The National Institute for Environmental Studies began testing the use of “mdx” in May 2023, and built an equivalent computational infrastructure on “mdx”, targeting numerical models running in the institute’s on-premises environment. The numerical model for verification is the global high-resolution atmospheric model NICAM. The goal is to stably operate at 160 parallel, including pre- and post-processing of numerical models performed on the on-premises side. To build a computational infrastructure to achieve this goal, the following tasks were performed:
- Design of standard virtual machine (VM) templates for use on “mdx”
- Construction of virtual machines according to the following roles:
|-- Login VM
|-- Authentication infrastructure VM
|-- Network File sharing VM
|-- MPI VM
`-- Web server VM
- Verification of numerical model operation
- Verification of operation and maintenance
- Verification of usage cost
The “mdx” management office provides virtual machine templates and recipes for building the required server functions through the official website. Since there are no cases like white papers that are operated as production environments, users need to verify themselves based on these documents. At this time, it is necessary to verify whether a system that complies with the information system usage regulations and functions that can be realized in the on-premises environment can be built. For functions that are lacking in official templates and documents, users must create their own virtual machine templates.
As a result of verification, the official template was adopted as MPI VMs, and for other VMs, we decided to create our own virtual machine templates. We focused on securing documents on the user side, including design documents and procedures, especially for parts related to operation and maintenance, including security. This ensures that you are not vendor-locked into “mdx” and ensures extensibility, such as deployment to other virtual infrastructures.
On the other hand, for the MPI VM, it is closely related to the hardware and storage services on “mdx”, and operational reliability is required. Therefore, we adopted an official template that is easy to get support from the “mdx” management office.
We built the computational environment shown in Figure 1 and were able to operate a model that was parallelized to 160. Currently, we are operating as if it were a production, verifying operating costs and verifying the expansion functions needed in the future.
In the presentation, we will introduce the details of the design of the virtual machine obtained as a result of this case and the advantages and disadvantages of migrating to “mdx”.
The National Institute for Environmental Studies began testing the use of “mdx” in May 2023, and built an equivalent computational infrastructure on “mdx”, targeting numerical models running in the institute’s on-premises environment. The numerical model for verification is the global high-resolution atmospheric model NICAM. The goal is to stably operate at 160 parallel, including pre- and post-processing of numerical models performed on the on-premises side. To build a computational infrastructure to achieve this goal, the following tasks were performed:
- Design of standard virtual machine (VM) templates for use on “mdx”
- Construction of virtual machines according to the following roles:
|-- Login VM
|-- Authentication infrastructure VM
|-- Network File sharing VM
|-- MPI VM
`-- Web server VM
- Verification of numerical model operation
- Verification of operation and maintenance
- Verification of usage cost
The “mdx” management office provides virtual machine templates and recipes for building the required server functions through the official website. Since there are no cases like white papers that are operated as production environments, users need to verify themselves based on these documents. At this time, it is necessary to verify whether a system that complies with the information system usage regulations and functions that can be realized in the on-premises environment can be built. For functions that are lacking in official templates and documents, users must create their own virtual machine templates.
As a result of verification, the official template was adopted as MPI VMs, and for other VMs, we decided to create our own virtual machine templates. We focused on securing documents on the user side, including design documents and procedures, especially for parts related to operation and maintenance, including security. This ensures that you are not vendor-locked into “mdx” and ensures extensibility, such as deployment to other virtual infrastructures.
On the other hand, for the MPI VM, it is closely related to the hardware and storage services on “mdx”, and operational reliability is required. Therefore, we adopted an official template that is easy to get support from the “mdx” management office.
We built the computational environment shown in Figure 1 and were able to operate a model that was parallelized to 160. Currently, we are operating as if it were a production, verifying operating costs and verifying the expansion functions needed in the future.
In the presentation, we will introduce the details of the design of the virtual machine obtained as a result of this case and the advantages and disadvantages of migrating to “mdx”.