1 A First Step Towards the Resolution of The Starvation Problem In Multi-Point-to-Point ICRCNs Wissam Fawaz, Ribal Atallah and Maurice Khabbaz Abstract—This letter revolves around an Intermittently Con- nected Roadside Communication Network (ICRCN) scenario consisting of isolated source Stationary Roadside Units (SRUs) exploiting mobile smart vehicles as store-carry-forward data relays to a destination SRU. In this case, it is observed that a subset of these source SRUs may suffer from a significant starvation problem. In this letter, first, an Markov Decision Process (MDP) framework is established for the purpose of identifying a suitable Bulk Release Decision Policy (BRDP). Second, BRDP is implemented within a Starvation Mitigation and Delay-Minimal (SMDM) bundle delivery scheme. Extensive simulations are conducted for the purpose of: a) quantifying the severity of the starvation experienced by the downstream SRUs and b) gauging the merit of the proposed SMDM scheme through its ability to jointly mitigate starvation and achieve end-to-end delay minimal bundle delivery to the destination SRU. Index Terms—ICRCN, SRU, Performance Evaluation, MDP. I. I NTRODUCTION T HE utilization of the transportation infrastructure as a means for establishing connectivity among isolated Sta- tionary Roadside Units (SRUs) represents an emerging terres- trial application of the Disruption-Tolerant Networking (DTN) paradigm, [1]. Recently, this application has gained significant momentum. This is especially true since it has been proven to be an effective and cost-minimal solution for bringing digital connectivity to rural areas, where the cost of setting up a networking infrastructure can be elevated [2]. This letter considers a networking scenario such as the one depicted in Figure 1 which consists of three SRUs, namely: a) S 1 and S 2 being two source SRUs and b) D being a destination SRU. All three SRUs are deployed along a one-dimensional and uninterrupted roadway segment. Each one of these three SRUs is located outside the respective coverage ranges of the two others and hence the three SRUs cannot directly communicate with one another. Furthermore, only D is connected to the Internet through minimal networking infrastructure. The two source SRUs S 1 and S 2 are completely isolated. In the absence of networking infrastructure connecting S 1 and S 2 to D, mobile vehicles equipped with computerized modules, finite buffers and wireless communication devices serve as store-carry-forward data carriers from both of S 1 and S 2 to D. Such a networking scenario belongs to the class of Intermittently Connected Roadside Communication Networks (ICRCNs). Upon entering the communication range 1 of either W. Fawaz and R. Atallah are with the ECE department of the Lebanese American University. E-Mail: {wissam.fawaz, ribal.atallah}@lau.edu.lb M. Khabbaz is with the ECE department of the Notre-Dame University. E-Mail: maurice.khabbaz@gmail.com 1 In the sequel, the event of a vehicle entering the communication range of an SRU is referred to as a vehicle arrival to that SRU. D S2 S1 d 2 d 1 Wireless Access Point SRU City Metropolitan Area Network THE INTERNET Fig. 1. ICRCN sub-network scenario. one of the two source SRUs S 1 or S 2 , a vehicle presents to that SRU a bundle 2 release opportunity. In the context of a point-to-point ICRCN with a single source SRU, the work of [2] aimed at determining the suitability of such an arising opportunity in terms of the minimization of the average end- to-end delivery delay of a singly released bundle. In a similar context, the authors of [4] addressed the limitations of [2] and demonstrated how the release of a subset of the bundles 3 queueing in a source SRU’s buffer results in a significant performance improvement. Throughout this present work, it is observed that for the duration of their journey over the roadway segment illustrated in Figure 1, vehicles will first enter the coverage range of S 1 and then, on their way to D, they will pass by S 2 . In turn, each of S 1 and S 2 will opportunistically attempt to load the arriving vehicles with as many as possible of their respective data bundles, if available. However, since the vehicles’ buffer sizes are finite, the middle source SRU, S 2 , is highly likely to suffer from a bundle release restriction or even a denial of bundle release. This is particularly true since the buffer of an arriving vehicle to S 2 might have been considerably loaded (ultimately fully exhausted) with S 1 ’s bundles. Typically, a node such as S 2 is referred to as a starving node since bundles would rapidly accumulate in S 2 ’s buffer and thus would forcefully experience excessive queueing delays. What distinguishes this letter from the work of [2], [4] is that it aims at both highlighting and mitigating the severity of the starvation that a downstream SRU such as S 2 may suffer from. 2 Data and control signals are combined in a single atomic entity, called bundle, that is transmitted across a DTN-based ICRCN, [1]. 3 Thereafter, a subset/group of bundles is referred to as a bulk of bundles.