Academic Center for Computing and Media Studies (ACCMS) offers high-performance computing services using its supercomputers. This service is widely used for a broad range of applications requiring massive computing power, such as large-scale simulations and visualization, by researchers across Japan, including those at Kyoto University.
　Data-driven research, on the other hand, requires diverse computing needs not covered by our high-performance computing services. These include the collection and management of data from various sources like laboratory equipment, sensors, and the Internet, as well as the analysis of collected data and the publication of datasets. Furthermore, there are scenarios where current supercomputers may not be available, such as when the analysis of sensitive data requires an isolated environment.
　This article introduces the services provided by ACCMS to support data-driven research.

mdx: A Platform for Building Data-Empowered Society

　mdx is an IaaS (Infrastructure as a Service) cloud platform designed for academia, which provides a private computing environment isolated for each research project using virtualization technology. On mdx, users can configure separate virtual networks for each research project, build an isolated, high-performance computing environment by connecting high-performance virtual machines (including GPUs), and leverage large-scale, high-speed storage (Figure 1). Users can configure these computing resources via a web-based self-service portal. Furthermore, by integrating with L2VPN services of SINET (the Japanese academic network operated by the National Institute of Informatics), users may be able to connect their laboratory's networks or sensors in remote locations to the same virtual network (Please consult with ACCMS or the operation team for each system).
　mdx is jointly operated by a consortium of 11 institutions, including ACCMS. mdx I is installed at the Information Technology Center, The University of Tokyo, and mdx II is installed at the D3 Center, The University of Osaka. mdx is equipped with GPUs, which are currently in very high demand. mdx I has NVIDIA Tesla A100, and mdx II has NVIDIA H200 SXM. More than several hundred terabytes (TB) of storage is also available for use. For further details, please refer to https://mdx.jp/.

Image

Figure 1 Overview of mdx (Cited from the press release published on March 9, 2021)

Edge Computing Infrastructure

　Various computing resources are available to researchers for data-driven research, such as our high-performance computing services, mdx, research data storage services (RDM Drive and RDM Object Storage) provided by the Institute for Information Management and Communication (IIMC).
　In actual research scenarios, these systems will be used in combination with laboratory instruments, sensors, and computers in laboratories and remote locations. For instance, imagine a workflow where sensitive data is collected from laboratory instruments or sensors into RDM Drive or RDM Object Storage, while large-scale analysis is performed on a supercomputer. We would need to bridge the isolated network connected to the laboratory instruments and sensors with RDM Drive or RDM Object Storage, and to transfer the data from RDM Drive or RDM Object Storage to the storage of supercomputer while removing sensitive data. However, no dedicated service has been available to support such cross-system bridging workloads.
　To support such workloads that bridge multiple services, ACCMS introduced an edge computing infrastructure at the end of FY2024, as part of the initiative “Establishment of the Data Operation Support Platform Center for Cross-sectional Data Drive to Generate Research DX” (FY2023?FY2027, by IIMC, ACCMS and Kyoto University Library Network) and its related project, “mdx Linkage and Data-Driven Infrastructure”.
　This system (Edge Computing Infrastructure) provides a container execution platform. Containers are a technology that achieves highly portable software execution environments by bundling the application and necessary files into a single container image, and the use of containers has been expanding rapidly in recent years. Container images are publicly available for a wide variety of software, and users are expected to save the effort of software setup that is typically necessary when using a virtual machine. This system supports diverse container execution patterns, such as continuous execution, periodic execution, or one-time execution.
　This system adopts Kubernetes, the de facto standard for container orchestration tools, and provides users with access to the Kubernetes API. Therefore, researchers can use this system not only for running “bridging” workloads but also as a general container execution environment (Figure 2). In addition, a wealth of expertise and documentation is readily available on the web.
　This system is connected to Kyoto University’s campus network, KUINS, at 200 Gbps, allowing containers running on the system to access other systems at high speed. Furthermore, the system is designed to enable communication with private networks connected to laboratory instruments or sensors by extending KUINS VLANs into the system (Please consult ACCMS for details). The power supply of this system is backed up by UPS and power generators to avoid service downtime during power failures, and it operates 24 hours a day, 365 days a year, except during maintenance or outages.
　As of the writing of this article (December 2025), this system is under preparation for service. We plan to launch the service following a pilot service scheduled for FY2026.
 

Image

Figure 2 Overview of Edge Computing Infrastructure 

（Daisuke Kotani：Academic Center for Computing and Media Studies, Kyoto University）