SFO: SubFlow Optimizer for MPTCP in SDN

Abstract

Multipath TCP (MPTCP) brings promising features of throughput improvement and load balancing by utilizing the multiple available paths. However, adding a subflow without awareness of the network state will result in wasting network resources and degrading MPTCP performance. Finding and operating the proper number of subflows are necessary to achieve the best performance of MPTCP. However, an MPTCP host does not have knowledge about the number of available paths in the network, and it may create the suboptimal number of subflows. This paper proposes SubFlow Optimizer (SFO) for MPTCP in SDN that estimates the Optimum Number of Subflows (ONS) and informs it to the MPTCP hosts. The estimated ONS reflects the number of optimal paths found out of the set of feasible paths between the server and client. Selecting such an optimal path for a new subflow improves its throughput whereas detecting ONS makes efficient use of the overall network. SFO also enables the quick negotiation of ONS between MPTCP hosts during its session establishment that reduces the time required for ONS negotiation down to one Round-Trip Time (RTT). Therefore, MPTCP hosts can immediately achieve the best performance without incremental try and catch of subflow creation. The results of performance evaluation show that SFO brings the significant throughput improvement to the overall MPTCP session compared with the cases of improper estimation of the available number of subflows.