Japan Reaches 775 Gb/s Storage Transfer, Takes 400G Test to Chicago

Conceptual photorealistic image of a Southeast Asian network technician handling fiber optic connections among data center racks in blue light; the previously approved BitcoinVersus.tech neon-green-on-black tag is copied once at bottom left.

Japanese researchers have pushed real storage workloads close to the limits of an 800 Gb/s optical connection, then brought a trans-Pacific follow-up to the SC26 supercomputing conference in Chicago. RIKEN, Japan’s National Institute of Informatics (NII), the University of Tokyo, NTT East and NTT West reported a 775 Gb/s peak between Kobe and Kashiwa, with rates exceeding 750 Gb/s for several minutes. The domestic measurements were announced on October 30, 2026. A separate end-to-end 400 Gb/s Japan–U.S. disk-transfer experiment was scheduled at SC26 in November, but a final measured international throughput result has not been independently established.

NTT’s official April 2026 explanation of IOWN and All-Photonics Networks. General infrastructure background, not footage of the October storage benchmark. Watch directly on YouTube.

Why 775 Gb/s is more interesting than an 800G port

A network interface’s advertised bandwidth does not establish how quickly a large scientific dataset actually moves between storage systems. Disks, software, packet overhead and transport protocols can all limit the result. The Japanese team connected the RIKEN Center for Computational Science in Kobe and the University of Tokyo’s Kashiwa campus over NTT’s All-Photonics Network, or APN, infrastructure. Each research site fed two 400 Gb/s connections into the 800 Gb/s intercity optical link.

The test ran two practical storage applications simultaneously: the Gfarm-based HPCI Shared Storage test environment and NII’s MMCFTP file-transfer tool. The reported 775 Gb/s was their combined peak traffic, not a claim that either application independently sustained 775 Gb/s. Measured against the nominal 800 Gb/s line, 775 ÷ 800 = 96.875%, rounded to approximately 97% utilization. The group also observed transfer speeds above 750 Gb/s for several minutes rather than continuously throughout the entire testing session.

What makes MMCFTP different?

MMCFTP stands for Massively Multi-Connection File Transfer Protocol. NII’s software runs many TCP connections in parallel and adjusts their number to keep data moving at a selected rate across long distances. The protocol can also use multiple network paths. For the Japanese storage experiment, the team deployed servers populated with NVMe solid-state drives; each used one 400GbE connection and four 100GbE interfaces to its local switching fabric.

The results demonstrate how software and storage engineering can help exploit fast fiber, not merely how fast an optical transceiver can signal. Related BitcoinVersus.Tech coverage explains optical transceivers and their interfaces and a 64-port 800G data-center switch. A 51.2 Tb/s switch fabric and a 775 Gb/s end-to-end application transfer measure different parts of an infrastructure system.

Techzine’s August 2026 interview with an NTT representative on All-Photonics Networks and data-center interconnects. Technology context, not a certification of the 775 Gb/s result. Watch directly on YouTube.

The next test crosses the Pacific

At SC26, held November 15–20 in Chicago, NII and its partners registered a disk-to-disk trial on a single end-to-end 400 Gb/s path between Japan and the United States. The project appears in the official SC26 Network Research Exhibition under demonstration NRE-109. The design builds on Japan’s SINET6 international network and U.S. research connectivity; earlier MMCFTP trials frequently used aggregated 100G circuits or memory-to-memory tests to avoid disk input/output limits.

The new experiment addresses a harder question: can actual storage endpoints exploit a high-rate international network rather than just move synthetic traffic between memory buffers? Conference registration and demonstration plans are documented, but the public material checked does not supply an independently verified final 400 Gb/s disk-to-disk throughput result. The 400 Gb/s figure identifies the connection capacity, not a measured achieved transfer rate.

What the results mean for AI and science

Large telescopes, particle-physics experiments and AI research produce datasets that must move among remote computing sites. Faster file movement can shorten delays before analysis and reduce the need to manually ship drives, but no transfer benchmark establishes application cost, power savings or availability on every production network. NTT also reported roughly 20% lower measured latency for the APN test environment compared with the SINET connection in the same evaluation; the comparison is specific to those routes and equipment.

Other projects show the growing scale of optical links. Our reporting covers an 800G coherent optical trial over 2,104 km and SC26’s volunteer-built research network. The next useful disclosure for the Japan–U.S. experiment is a time-series trace of measured application throughput, storage I/O and network loss over the full international path.

Primary sources: RIKEN’s October 30 results; NII’s experiment report; NII’s November 16 SC26 announcement; SC26 accepted network research demonstrations.

We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support to help further secure the integrity of our research initiative, please scan the QR code below.

BitcoinVersus.Tech voluntary support QR code for independent research

Leave a Reply