A novel hybrid forwarding strategy for content delivery in wireless information-centric networks

Abstract

Information-centric networking (ICN) is a promising solution for content delivery in multi-hop wireless networks. In these environments, the ICN forwarding strategies typically rely on broadcasting, which facilitates content distribution by taking advantage of the shared medium, but brings about undesirable side effects. To avoid the broadcast-related issues of packet redundancy and unreliability due its unacknowledged mode, a few recent proposals advocated unicasting as the communication mode to resort to, once the content provider has been discovered. However, unicasting may suffer from connectivity breakages due to node mobility and harsh propagation conditions, and, generally, limits the data sharing capability of the wireless medium.

In this paper, we design a robust ICN-based forwarding strategy, called Ad Hoc Dynamic Unicast (ADU), that wisely harnesses unicast and broadcast primitives on top of IEEE 802.11-based wireless networks. ADU relies on unicasting for content dissemination after the provider discovery, and promptly falls back to broadcast to find a new content provider in case of a link failure notified by the Medium Access Control (MAC) layer. Extensive simulations in two different multi-hop wireless scenarios, i.e., mobile and vehicular ad hoc networks, under different load settings, assess the ADU performance against benchmark ICN forwarding schemes, representative of broadcast-based and unicast-based solutions. Results show that ADU outperforms them in all circumstances, proving its better responsiveness to link failures, which guarantees a shorter content retrieval delay and a lower message overhead and energy consumption.

Introduction

Research in the area of multi-hop wireless ad hoc networking is today revamping pushed by a more pragmatic development strategy [1]. Many scenarios such as public city-wide dedicated services (e.g., video-surveillance or enhanced transportation systems), environmental monitoring through sensing systems, and post-disaster management, ask for easy and quick, highly scalable and cost-effective network deployments, with the aim of extending, replacing and/or offloading the infrastructure. There, either densely or sparsely deployed purpose-built devices (e.g., sensors, access points, road-side units), but also smartphones carried by pedestrians [2], [3], and other networked objects like cars [4], may play an active role in the generation, sharing, dissemination and retrieval of contents.

Contents may span from small sized data (e.g., measurements retrieved from sensors) to medium-to-large files (e.g., city maps, pictures of a road segment) and may exhibit different popularity levels and time- or spatial-relevance. For instance, road traffic information are public utility (highly popular) locally-relevant data, the time validity of which lasts a few seconds or minutes.

Recent literature articles argued in favour of Information Centric Networking (ICN) [5] as a key paradigm to facilitate content retrieval in wireless ad hoc networks [4], [6]. The main ideas behind this paradigm, i.e., (i) content names used at the network layer for data retrieval, instead of end-host network addresses, (ii) interaction models decoupling sender (a.k.a. provider) and receiver (a.k.a. consumer) and supporting asynchronous communications, and (iii) in-network caching, well suit environments where connectivity with a specific end-host is difficult to set-up and maintain, because of the error-prone radio channel, the mobility of nodes or their battery-powered nature.

In the majority of ICN solutions for wireless environments such as mobile ad hoc networks (MANETs) [7], [8], vehicular ad hoc networks (VANETs) [9], [10], wireless sensor networks (WSNs) and, at a larger scale, the Internet of Things (IoT) [11], [12], [13], the requests for a named content are typically sent in broadcast to maximize the possibility of finding a content by taking advantage from packets overhearing and distributed in-network caching. However, broadcasting - if not wisely controlled - can lead to congestion and high packet redundancy, with consequent inefficient bandwidth usage and waste of devices’ resources. Therefore, packet suppression techniques and/or further signalling have been proposed [8] to keep redundancy under control.

Meanwhile, few recent works argued in favour of unicasting, once content providers have been discovered through the initial request broadcasting [14], [15], [16]. Unicasting has the advantage of leveraging mechanisms enforced at layer-2 (acknowledgments and retransmissions) to guarantee reliability. Moreover, clearly, packet redundancy is avoided by letting each node, on the path towards a given content provider, address a single neighbour as the next-hop forwarder.

Solutions relying on unicasting are still at their preliminary development stage and need to be further improved and evaluated. Current schemes mainly consider a request retransmission timeout (RTO) maintained by the consumer at the ICN layer to trigger a recovery from losses along the path. When the RTO expires and no data has been received, the consumer restarts the content provider discovery by transmitting a request in broadcast. The detection of a link failure (which is likely to occur in a dynamic network environment) only upon the RTO expiration can penalize the content retrieval delay. Mechanisms are required, instead, that quickly adapt to time-varying topology/channel conditions. In order to advance the state-of-the-art, in this paper we provide the following key contributions.

• We further elaborate on the concept of unicast forwarding and take a step forward by conceiving a solution called Ad-Hoc Dynamic Unicast (ADU), which wisely combines unicast and broadcast and is highly tied to the underlying Medium Access Control (MAC) layer dynamics. Focus is on IEEE 802.11 [17], which is the predominant access technology for wireless ad hoc networks. ADU proposes a novel MAC-driven recovery protocol, built upon the solution in [15], which is able to promptly revert to broadcast in case of a link breakage reported by the MAC layer. Such a mechanism is implemented not only by consumers, but also at intermediate nodes, to recover more quickly. In addition, ADU foresees that consumer applications enforce retransmissions of Interest packets at the expiration of the RTO. The RTO computation follows the Transmission Control Protocol (TCP) procedure, also leveraged in previous ICN literature [18], [19], with proper workarounds aimed to increase the stability and reliability.

• To give insights into the benefits of ADU we compare it against two reference schemes, respectively, representative of (i) an enhanced broadcast-based solution [8], which tracks the identifier of a discovered content provider, and (ii) a unicast-based one [15], which ADU builds upon. We perform a comprehensive simulation study, through ndnSIM [20], in two different scenarios, characterizing a MANET and a VANET, respectively, under varying traffic load settings. The two scenarios feature different dynamics, e.g., in terms of mobility and propagation settings. Thus, the ability to support both, while outperforming the benchmarking solutions, underlines the flexibility of the designed ADU. It can leverage the benefits of unicast forwarding in reducing message redundancy and overhead, but it also guarantees short content retrieval delay since it promptly switches to broadcast when the path towards the content provider breaks.

The remainder of this paper is organized as follows. Section 2 provides a short overview of the ICN paradigm, while Section 3 discusses its applicability in IEEE 802.11-based wireless ad hoc networks and scans the state of the art, with special focus on representative broadcast and unicast-based forwarding schemes. Section 4 shortly discusses the reliability issue in ICN wireless ad hoc networks, by reviewing the related literature solutions. Section 5 presents the proposed ADU protocol. The simulation framework and settings are presented in Section 6, before discussing the results of the conducted simulation campaigns in Sections 7 and 8. Section 9 summarizes the main findings of the analysis, before wrapping up conclusive remarks and giving hints on future works.