An Analysis of the Impact of Out-of-Order Recovery Algorithms on MPTCP Throughput

In this paper we evaluate and compare the end-to-end performance of different multipath TCP (MPTCP) congestion controllers when run in conjunction with different TCP packet reordering recovery algorithms. The paper answers the following questions: what is the impact of out-of-order events on the end-to-end throughput when using MPTCP, how do out-of-order recovery algorithms that have been proposed for single-path TCP perform with multi-path TCP, and how sensitive this performance is against the delay difference between the paths used. The paper compares three different MPTCP congestion control algorithms used in conjunction with four current TCP packet reordering solutions: D-SACK, Eifel, TCP-DOOR, and F-RTO. Simulation results show that whilst TCP-DOOR and, second, D-SACK perform generally better across all congestion control MPTCP implementations, the choice of packet reordering algorithm is not always fixed and straightforward - when MPTCP is used with some form of coupled congestion control the performance degrades towards that of single-path usage when the delay difference of the paths is over 200ms. The paper identifies combinations of congestion control and packet reordering algorithms that give better aggregate throughput performance for different path delay differences.