AIST’s Green Cloud Infrastructure expands the existing AI Bridging Cloud Infrastructure (ABCI), which was provided by Fujitsu and has been in operation since August 1, 2018, and is scheduled to start operations in early 2021. More than 1,000 users, from AI start-ups to electronics manufacturers, have been conducting research and development using the current system. Faced with rapid growth in demand for AI research and development in recent years in industry, government and academia, AIST has decided to add the “AI Bridging Green Cloud Infrastructure” in collaboration with ABCI. The concept is to operate both systems in an integrated manner.
The new storage environment, which will be used on both systems, consists of an 11.2 petabyte high-capacity storage system based on DDN’s EXAScaler parallel file system solution. EXAScaler is based on open-source Lustre, which has been adopted by many supercomputer systems and large-scale AI computing infrastructure projects, both domestically and internationally.
The first storage vendor to widely adopt the open-source Lustre file system, DDN acquired the Lustre development division from Intel in 2018 and has since focused on optimizing Lustre for modern applications in data analytics and AI. In addition to maintaining the Lustre source code tree, DDN is now also the main contributor to the Lustre source code.
The DDN storage system selected for the ‘AI Bridging Green Cloud Infrastructure’ comprises three DDN 400NVX units with 405.5 terabytes of effective data capacity, and uses state-of-the-art NVMe SSDs, as well as DDN ES7990X™ units with large-capacity hard disks for a total 10.8 petabytes of usable data capacity. The high-speed data pool will deliver an effective write throughput of 105GB/s, with read throughput of over 120GB/s, and over 4 million random read file IOPs, while the capacity area is expected to deliver an effective throughput of at least 60GB/s.
“DDN’s EXAScaler is a proven solution that has been delivering high-performance file systems for large HPC environments for some time now. Over the past few years, DDN has invested heavily in development related to the Lustre file system, working closely with GPU vendors such as NVIDIA to optimize Lustre for a variety of data analytics, machine learning and AI workloads,” said Shuichi Ihara, principal engineer working mainly on performance improvements at Whamcloud, which is responsible for developing DDN’s Lustre-based EXAScaler file system solution. “We are honored to be selected by AIST as the main storage provider for Japan’s national AI infrastructure, built by AIST in collaboration with industry, academia and government. We are excited to be able to contribute to the development of AI research and development in Japan together with AIST and Fujitsu.”
