On the Fidelity of NS-3 Simulations of Wireless Multipath TCP Connections

The multipath transmission control protocol (MPTCP) is considered a promising wireless multihoming solution, and the 3rd generation partnership project (3GPP) includes it as a standard feature in the fifth-generation (5G) networks. Currently, ns-3 (Network Simulator-3) is widely used to evaluate the performance of wireless networks and protocols, including the emerging MPTCP protocol. This paper investigates the fidelity of the Linux kernel implementation of MPTCP in the ns-3 direct code execution module. The fidelity of MPTCP simulation is tested by comparing its performance with a real Linux stack implementation of MPTCP using a hardware testbed for two different setups. One setup emulates the existence of a bottleneck link between the sending and receiving networks, whereas the other setup does not have such a bottleneck. The fidelity of ns-3's simulation is tested for four congestion control algorithms, namely Cubic, linked-increases algorithm (LIA), opportunistic LIA (OLIA) and wVegas for relatively short and long data flows. It is found that the uplink MPTCP throughput performance exhibited by the ns-3 simulator matches the hardware testbed results only if the flows are long-lived and share no common bottleneck link. Likewise, the MPTCP throughput achieved during a downlink scenario using the ns-3 simulator and the hardware testbed are close to each other across all algorithms except wVegas regardless of the flow size if there is no bottleneck link. Moreover, it is observed that the impact of LTE handover on MPTCP throughput is less significant in the simulator than the real hardware testbed, and it is setup-dependent.