[M-GI31] Earth and planetary informatics and data utilization

[MGI31-P12] HpFP: development of high-speed data transfer applications (5): Development of HpFP version 2

Yasunori Kakizawa¹, Praphan Pavarangkoon³, Ken T. Murata², *Takamichi Mizuhara¹, Keiichiro Fukazawa⁴ (1.CLEALINK TECHNOLOGY CO., LTD., 2.National Institute of Information and Communications Technology, 3.School of Information Technology King Mongkut's Institute of Technology Ladkrabang, 4.Research Institute for Humanity and Nature)

Keywords:HpFP, HCP, High-speed data transfer, Network protocol

For LFNs (Long-Fat Networks) with 10 Gbps or more and satellite networks with large latency, a variety of TCP-based protocols have been proposed which show high performance on large latency network conditions. However, such TCP protocols are essentially unable to achieve large bandwidth on high latency networks accompanied with packet-losses that are inevitable on practical LFNs or satellite networks. To overcome this issue, we designed a new data transfer protocol on TCP/IP transport layer built on top of UDP: High-performance and Flexible Protocol (HpFP). It constantly monitors latency (RTT) and packet losses and conducts rate control and retransmission control based on them to enable higher bandwidth data transfer than 10 Gbps even on packet-loss conditions over LFNs. The basic concepts are addressed and protocol design of the HpFP are discussed. An application (named as Archaea tools) to transfer many data files in small size (e.g., 1MB x 10000 files) with almost wire-rate bandwidth (10 Gbps) is also addressed. In this study we develop a tool to bind multiple connection of HpFP aiming at higher throughput on a broadband network faster than 100Gbps. In the near future big data transfer with high speed is expected on LFNs such as SINET5.
We are currently developing various modes for HpFP, including fair mode that emphasizes fairness, super-aggressive mode that maximizes performance on dedicated networks, cubic mode that is almost the same implementation as TCP cubic, and modest mode that yields the maximum amount of bandwidth when there is other traffic. In our talk, we introduce mechanisms of each mode and report on the results of laboratory experiments and experiments in a real JGN/SINET network environment.
This work is supported by "Joint Usage/Research Center for Interdisciplinary Large-scale Information Infrastructures" and "High Performance Computing Infrastructure" in Japan (Project ID: jh230057 and jh240077) and by JSPS KAKENHI Grant Number JP22H01318 and by Council for Science, Technology and Innovation (CSTI).