A Memory-Efficient Transmission Scheme for Multi-Homed Internet-of-Things (IoT) Devices.

Internet-of-things (IoT) is a wide spreading technique that enables intelligence to the everyday objects, however, IoT devices are limited in computation and memory space due to their small physical sizes. As a result, IoT applications generally connect to the remote cloud that provides high computation and large storage. To enhance this communication, some IoT devices are equipped with multiple networks, e.g., cellular or Wi-Fi, by using Multipath TCP (MPTCP). However, MPTCP requires large buffer memory space compared to the legacy TCP, which is problematic for low-memory IoT devices. This paper proposes a new MPTCP scheme that leverages the multi-homed feature of low-memory IoT devices. Our design utilizes an application-level distributor that transmits packets to each MPTCP socket at each endpoint of the data. This scheme cleverly avoids the buffer blocking problem, while still maintaining the benefits of multi-homing. The main contribution of our paper is three-fold. First, our proposal achieves the benefits of multipath while avoiding buffer blocking due to out-of-order packets. Second, since our scheme utilizes the original MPTCP and modifies only the application level, it can be deployed more easily to the legacy systems. Finally, our experimental results, conducted on a Linux testbed and real-world cellular/Wi-Fi, show that the proposed scheme requires only half or less memory to achieve the performance of MPTCP.