NII, National Institute of Informatics (NII, Director General: Masaru Kitsuragawa, Chiyoda-ku, Tokyo) and NICT, National Institute of Information and Communications Technology (NICT, President: Hideyuki Tokuda, Koganei, Tokyo) has conducted data transfer experiments between Japan and the United States using "MMCFTP" file transfer protocol (Massively Multi-Connection File Transfer Protocol) developed by NII, and successfully transferred 10 TB data stably. It is 81 Gbps greater than 150 Gbps recorded last year, and it reported to be as the "world's fastest" (one to one server transfer rate) as intercontinental Long-Distance Data Transmission.
300Gbps experiment network between Japan and the US
In this experiment, we have extended the experimental network environment of "JGN", the research and development testbed network operated by NICT, to the NICT booth in SC 17 venue in Denver, USA. We have constructed three independent experimental networks consist of JGN domestic network in Japan, JGN Asia 100Gbps network (Tokyo - Hong Kong - Singapore) which began its operation in November, SINET 5 provided by NII, and many academic research networks in Japan and abroad. Cases like this, when communication lines are used at the near performance limit in the international large capacity data transmission experiments, setting adjustments in accordance with experimental requirements at various communication equipment in the networks are required. Therefore, strong operation coordination among the organizations that operate each of the networks is necessary. NICT has made a significant contribution to this success by providing a wide range of experience of demonstration experiments using over 100Gbps bandwidth and strong relationship with oversea organizations.
Structure of test system
We installed transmitters and receivers (general purpose server) for MMCFTP in NICT 's JGN Access Point in Tokyo and NICT booth in SC 17 venue in Denver. In addition to "SINET 5" operated by NII and "JGN" operated by NICT, we also had the cooperation of various international networks such as “Internet 2” the experimental network consists of the US academic network, "SingAREN" academic network in Singapore, "TransPAC" operated by Indiana University, "PacificWave" academic network in the US, also Japan's "WIDE project" and SC17 venue network "SCinet" made this experiment possible.
Test Results
The test was conducted in the form of transferring data to Japan from the venue of "SC17" the international conference held in Denver, USA from 12 to 17 November 2017. The effective data transfer speed (goodput) for 10 TB was ranged from 224.9Gbps (5 minutes 55 seconds transfer time) to 231.3Gbps (5 minutes 45 seconds transfer time). 10 TB data is equivalent to 400 typical Blu-ray discs with 25 GB data capacity, or about 1200 hours of terrestrial digital broadcasting quality data. This enormous volume of data is transferred from the US to Japan in less than 6 minutes.
In this experiment, we used a total of 300Gbps experimental channels between the US and Japan consist of a 100Gbps channel via Singapore in addition to two 100Gbps channels directly connected between Japan and the US. We used as a part of the channel via Singapore, JGN/SingAREN Asia 100Gbps lines which NICT and SingAREN have jointly established in November 2017. The round-trip delay time (indicator of communication distance) of each channel was 125 milliseconds via Seattle, 122 milliseconds via Los Angeles and 288 milliseconds via Singapore. The channel via Singapore was more than 2 times longer than the other channels. When multiple channels with different communication distances are used at the same time, usually congestion occurs because of overloaded traffic flows in short-distance channels. When high-capacity high-speed data transmission test is conducted in an environment where normal academic traffic also flows, such congestion usually has a bad impact on the normal traffic and must be avoided.
MMCFP resolves this problem with its connection distribution control function. MMCFTP can specify the ratio of the number of TCP connections to be assigned to each path when data is transferred using multiple channels. In this experiment, we allocated TCP connection at a ratio of 1: 1: 4 via Seattle, via Los Angeles and via Singapore. As a result, the difference in the TCP connection speed due to the communication distance is neutralized by the difference in the number of connections, and the data was transferred through channels at 80.5Gbps via Seattle, 80.8Gbps via Los Angeles and 70.6Gbps via Singapore, while the remaining bandwidth is left for the normal academic traffic and achieved the transmission rate total of 231.3Gbps.
