A large amount of data needs to be transferred from one site to another as fast as possible in the computational science fields. To achieve high-speed data transfer in widely-distributed environments, many applications utilize multiple TCP streams. Using multiple TCP streams in parallel can improve aggregate bandwidth by mitigating the negative effects of packet loss and the slow start mechanism of TCP. However, since multiple TCP streams of applications are usually routed according to the default IP routing protocol, only a single shortest path among the multiple paths can be utilized for the data transfer. This research proposes a multipath controller that increases the performance of data transfer by leveraging multiple paths simultaneously for parallel TCP streams.
For this purpose, we utilize the Software-Defined Networking (SDN) technology and its implementation, OpenFlow. Furthermore, we proposed a prediction model to determine optimal numbers of parallel TCP streams to be assigned for each path according to its own network condition. This thesis presents the design and implementation of the proposed system. As a case study, we applied our proposed system on GridFTP and evaluated the performance improvement in both a virtual and a real global-scale environment. The results demonstrate that our proposed system accelerates the data transfer of GridFTP. In our real global-scale environment, our experimental results show the practicality of our proposal and indicate that our proposed method achieved the performance that is close to the physical limitation of the hardware.