- 著者
-
Chawanat Nakasan
Kohei Ichikawa
Putchong Uthayopas
- 出版者
- 一般社団法人情報処理学会
- 雑誌
- 研究報告ハイパフォーマンスコンピューティング(HPC)
- 巻号頁・発行日
- vol.2014, no.30, pp.1-6, 2014-07-21
This paper discusses the use of Multipath TCP (MPTCP), which is a TCP extension that allows multiple TCP flows to be associated to one application-layer logical connection, coupled with OpenFlow traffic engineering in a single stack to provide a comprehensive multipathing solution, with OpenFlow providing optimal path sets while MPTCP utilizing them. This design should be able to maximize bandwidth and network path utilization by allowing hosts to take advantage of presently-unused paths. Design of the testbed, also to be used by our research group in future projects, is also discussed in this paper. Finally, we discuss evaluation of network performance when using multiple paths, as well as concerns raised by our work. In summary, our system functioned as expected and provided feasible performance in small virtual network. This design should be scalable to benefit distributed file storage systems, data-intensive services, or any high-performance computing systems.This paper discusses the use of Multipath TCP (MPTCP), which is a TCP extension that allows multiple TCP flows to be associated to one application-layer logical connection, coupled with OpenFlow traffic engineering in a single stack to provide a comprehensive multipathing solution, with OpenFlow providing optimal path sets while MPTCP utilizing them. This design should be able to maximize bandwidth and network path utilization by allowing hosts to take advantage of presently-unused paths. Design of the testbed, also to be used by our research group in future projects, is also discussed in this paper. Finally, we discuss evaluation of network performance when using multiple paths, as well as concerns raised by our work. In summary, our system functioned as expected and provided feasible performance in small virtual network. This design should be scalable to benefit distributed file storage systems, data-intensive services, or any high-performance computing systems.