اینترنت اشیای بی سیم- هزینه کافی برای انتقال داده TCP در شرایط ارتباط رادیویی از دست رفته Wireless Internet of Things – Quantifying Cost to Transmit TCP Data in Lossy Radio Link Conditions

I. INTRODUCTION Many IoT applications rely on edge devices that are connected through radio links [1]. Information types in IoT devices typically fall into three classifications: 1) urgent real-time information that must be transacted at all costs (nominally less than 1kB); 2) normal communications between the device and backend host (nominally less than 10kB); and 3) over-the-air updates & maintenance communications (nominally 1MB or greater). For transferring data end-to-end, edge devices often make use of TCP. When IP packets cannot be delivered to the other end, TCP retransmissions are triggered to ensure that lost packets are in the end correctly received and delivered to the application layer in the right order. While being effective in achieving this goal, the TCP solution requires IP packets to be retransmitted over the radio channel. In addition, the TCP throughput performance may be compromised by a lossy radio link. This problem has been widely addressed in the literature [2], [3], [4], [5], [6], [7]. For most IoT applications, however, the edge device throughput is typically quite moderate and the TCP throughput performance does not represent an issue. More relevant to IoT applications is the inefficient utilization of resources caused by TCP retransmissions. For example, with some wireless ISPs, the subscriber is charged for IP packets being both transmitted and retransmitted. In other wireless ISPs, retransmitted packets represent a loss of revenue as the subscriber is only charged for each packet delivered successfully [8], [9]. From the device point of view, battery consumption is proportional to the amount of retransmitted packets, which may deplete unnecessary energy from the edge device. In summary, while TCP is a well-established packet retransmission mechanism for ensuring end-to-end data delivery, its inefficient resource utilization caused by lossy radio channel is perceived as a serious drawback from an IoT application point of view. In this paper, the authors argue that the first step towards addressing this drawback is to quantify the TCP “retransmission cost” caused by a lossy radio channel. While qualitatively it is straightforward to expect that a lossy channel is causing more TCP retransmissions over a given data transfer, the aim of this preliminary study is to quantify the effect of some of the radio channel performance indicators [10] on TCP packet retransmission. Based on the outcome of this preliminary study, the authors propose the creation of a functional layer between the application layer and the lower layers on the wireless edge device, referred to as “RF Fidelity Layer”. Its purpose is to provide real-time situational knowledge of the radio link layer to the application software, which in turn can “tune” the information flow for maximum efficiency and minimum billable data overage due to TCPdriven packet retransmissions.