In June, RIKEN shook the supercomputing world with its Arm-based, Fujitsu-built juggernaut: Fugaku. The system, which weighs in at 415.5 Linpack petaflops, topped the most recent Top500 list of the world’s most powerful publicly ranked supercomputers. At the time, its storage was outlined as a three-parter: 15.9 petabytes of NVMe storage, a Lustre-based global file system and a third, cloud-based storage option that went unnamed – until now. In a one-two punch of wins for the cloud provider, Oracle has announced that its Oracle Cloud offering is the service powering Fugaku’s cloud storage and highlighted the strong showing from Oracle Cloud Infrastructure (OCI) on the IO500 high-performance storage benchmark.
According to Oracle, the partnership between RIKEN and Oracle Cloud Infrastructure enables Fugaku’s users to work with the massive supercomputer through Japan’s Science Information Network (SINET) at ultra-high speeds. RIKEN has launched a joint project using this Oracle-provided cloud storage, aiming to leverage those cloud capabilities for “oversea adoption expansion efforts” in partnership with Altair, HPC Systems and the Foundation for Computational Science.
Oracle is also emphasizing its Cloud Infrastructure’s FastConnect tool, a dedicated network connection service that permits inbound and outbound bandwidth charges at no additional cost. “Research projects where data may extend well into petabytes can use Oracle Cloud Infrastructure’s high-performance computing and storage resources without having to worry about the massive data transfer cost and run the workloads with predictable low costs,” Oracle explained in a blog post.
In a separate blog post, Oracle shined a spotlight on its recent IO500 win. The IO500 benchmark, created by the Virtual Institute for IO in 2017, is a “comprehensive benchmark suite that enables comparison of high-performance storage systems” by measuring their read/write performance in a series of tasks. The most recent IO500 list placed Oracle Cloud Infrastructure as the 7th-fastest storage system.
“The vast majority of other systems on the list are specialized on-premises environments, including research supercomputers,” Oracle wrote. “We achieved 500 GB/s write IO throughput and 13.1 million IOPS in metadata performance with a high-performance parallel file system using 270 nodes in an HPC cluster network running the BeeGFS BeeOND file system.”
To provide a sense of scale, Oracle cited its own customer, the now-wildly popular video meeting app Zoom. Oracle says that Zoom moves about 7 PB of data per day to around 300 million users – and that the cluster used for its IO500 win could process that data locally in less than four hours.
“Oracle has gone a long way in targeting high-performance configurations and scalability with OCI,” commented Addison Snell, CEO of Intersect360 Research. “OCI has put up some big scores on the IO500, with a choice in parallel file system offerings. The relationship with RIKEN and Fugaku helps establish OCI as a serious presence in HPC.”