An Experiential Study of Multipath TCP Schedulers With the Comparison of Throughput, Latency, and Path Utilization: Multipath TCP Schedulers

Introduction

Advanced internet technology and modern telecommunication devices with multi-homing facilities encourage us to use multiple available interfaces concurrently for good quality of fast communication services. Multipath TCP(MPTCP) is IETF's (Ford et al., 2013) standard which provides a multipath facility, which is built over TCP. Unlike TCP, it allows the host to use many TCP sub-flows using different network addresses called multiple network interfaces for an application connection end to end, using multiple TCP sockets. MPTCP is capable to provide redundant connections as well as it gives the opportunity of bandwidth aggregation. MPTCP is the protocol, which is TCP's variant and built over it to handle TCP middleboxes and network devices, that work transparently with MPTCP as shown in Figure 1. It is also an application-agnostic. To avail the MPTCP facility, both the end hosts should be MPTCP capable, it can be implemented using the MPTCP_CAPABLE option in TCP. The connection of MPTCP's initial sub-flow is established the same as TCP, which is implemented by a three-way handshake. Later additional sub-flows can be added with ADD_ADDR and MP_JOIN commands. Unlike TCP, MPTCP's single connection uses many TCP sub-flows with different interfaces like Wi-Fi, LTE, 4G, 5G, etc. and all the sub-flows can be heterogeneous in nature with respect to bandwidth, path latency, loss factor, congestion window size, etc. MPTCP has below controls for network optimization.

• 1.

  Congestion Control: MPTCP uses a coupled congestion control technique for coordination of sub-flows congestion windows and queues. It goes through three phases of congestion slow start, congestion avoidance, and retransmission. It uses the principle of additive increase and multiplicative decrease (AIMD). It considers (congestion window) cwnd of all sub-flows.

Figure 1.

Multipath Tcp over TCP

• 2.

  Packet Scheduling: Packet scheduler of MPTCP is essential to distribute packets on different available sub-flows in efficient way and with least path delay. MPTCP uses Default/MinRTT (Minimum Round-Trip Time Scheduler) (Damon Wischik et al., 2011) as default scheduler in its Linux kernel built. This article covers theoretical and practical comparison of known MPTCP Schedulers like Default scheduler, Blest (Blocking estimation based MPTCP Scheduler) (Simone Ferlin et al., 2016), Roundrobin (C. Paasch et al., 2014) and redundant scheduler (Frommgen et al., 2016). Default/MinRTT scheduler maximizes the usage of sub-flow which has the least round-trip time (RTT) until the congestion window of the sub-flow is exhausted. It then switches to the next sub-flow with the least RTT. Blest scheduler tries to maximize sub-flow utilization by estimating the number of packets that can be sent on the least RTT sub-flow to compensate for the heterogeneity of two sub-flows. Round Robin scheduler uses multiple sub-flows alternatively irrespective of RTT and stops sending network traffic to sub-flow based on congestion window usage. The redundant scheduler uses all sub-flows for transmission of the same network traffic in a broadcast manner. In further sections of article motivation behind the work, literature survey and a comparison of different MPTCP schedulers are covered. Later sections show experimental setup and the result of experimentation.