Delay Tolerant Network Protocols in a Railway Network

Introduction

The article is an extended version of the paper “Delay Tolerant Network Potential in a Railway Network” (Tikhonov et al., 2020), published as part of the 2020 26th Conference of Open Innovations Association (FRUCT).

DTN (Delay Tolerant Network) as a part and the enhancement of the railway communication system is being analyzed in the article. Messages from a train to the server in the external fixed network and return messages to the train are transmitted over the radio-link. If the train is inside the network (the coverage area of the base stations, BS) then the message can be delivered almost instantly. If the train is out of the coverage then delivery is delayed by several minutes, tens of minutes, or even more than an hour. In the DTN case, if a train is offline (out of the coverage) during an attempt to send a message, then this message is saved. Then it may be forwarded to another train (oncoming or passing) when these trains are in mutual communication range. When the train reaches the coverage area, the message is sent – if it was not delivered yet through the forwarding chain by another train, which arrived in the coverage area earlier (see Figure 1).

Figure 1.

Example of two possible data delivery routes. Data is generated on the Mobile terminal 1 (black train) and could be delivered directly (dotted arrow) or partially via another Mobile terminal 2 (gray train) that is faster

The idea is to use DTN to reduce message delivery delays and/or increase the number of delivered messages. The paper estimates the potential improvement of these parameters due to DTN, and how close to this estimation specific DTN routing protocols can be. Computer simulation of the railway sections Tyumen-Surgut (low traffic) and Moscow-St. Petersburg (high traffic) is used as illustrative scenarios. At the same time, the relevant types of transmitted data are analyzed: telemetry (continuous data generation on the train with delivery to the external network), single user-messages (to the external network or to the train), dialogs (chains of responses to received messages) and message flows (generated both on trains and in the external network with a given intensity). The influence of the values of the main parameters is studied: the average speed of the response in the dialogs, the network coverage of the railway, the equipment of trains with satellite terminals, the key parameters of the routing algorithms.

Background

A system of movable mobile objects that could interact with each other is a mobile ad hoc network (Namiot, 2015). In the railway scenario not all objects can directly communicate with each other, links do not form a full graph. Intermediate relay nodes should store and forward packets (Ramanathan & Redi, 2002). So, data should be delay and disruption tolerant; therefore, this is a DTN case, with communication possibility determined by proximity (Namiot & Sneps-Sneppe, 2012).

There are a lot of works devoted to the effectiveness of DTN in various scenarios. The system tries to transmit the largest amount of data as quickly as possible. The general theoretical foundations of DTN are considered, for example, by Fall (2003). The main difficulties are discussed, and the basic terms and concepts are proposed. Jain et. al. (2004) propose a general categorization of possible routing protocols based on various prior knowledge (future predictions) divided into classes of “Oracles”. “First Contact” (FC) makes a random choice of the transmission route and does not require any a priori data. “Minimum Expected Delay” (MED) selects the best route according to average delivery delay statistics using the Dijkstra algorithm. “Earliest Delivery” (ED) uses current knowledge of upcoming delivery times. “Earliest Delivery with Local Queuing” and “Earliest Delivery with All Queuing” additionally balance the intermediate nodes load at a meeting or in the entire system, respectively. Theoretically, it is possible to minimize the total delay by using Linear Programming, but in practice, such a solution requires enormous computing resources and complete knowledge of the future. Therefore, it is necessary to rely on not optimal, but more realistic algorithms.