Experimental Verification of Massive Data Transfer for Super Intelligent Computing Service

With the development of super intelligent computing services, the requirement for massive data transfer (MDT) across long distance networks from edge data centers/enterprises to core data centers continues to increase. The network operators are strongly demanded to provide high efficiency MDT between the large-scales super intelligent computing centers. However, the conventional MDT scheme are achieved by manual express, which the valuable data suffer from security and damage risks. In this paper, we propose a low cost and highly efficient MDT scheme based on TCP for existing IP networks. To maximize the MDT rate and network bandwidth utilization, the TCP parameters are optimized. The performance of two congestion control algorithms, such as cubic and bbr, are studied in our experiment. The field experiments are also demonstrated to verify the MDT performance. The experiential results show that up to 9.18 Gbit/s MDT rate is achieved in a 1400 km field experiment with a 10 Gbit/s link rate. Moreover, it is worth noting that the proposed MDT scheme presents lower cost, higher efficiency, and flexibility for new supercomputing and intelligent computing services.