GITAR: Generic extension for Internet-of-Things ARchitectures enabling dynamic updates of network and application modules

Abstract

The Internet-of-Things (IoT) represents the third wave of computing innovation and relies on small, cheap and/or energy efficient devices, densely deployed in various spaces. Automatically managing, updating and upgrading the software on these devices, particularly the network stacks, with new, improved functionality is currently a major challenge. In this paper we propose GITAR, a generic extension for Internet-of-Things architectures, that enables dynamic application and network level upgrades in an efficient way. GITAR consists of four design concepts which can be applied to any operating system running on IoT/M2M devices. The proof of concept implementation in this paper uses the Contiki OS and the evaluation, based on analytical and experimental methods, shows that GITAR i) is up to 14% more efficient in terms of memory usage and ii) has less or similar run-time CPU overhead as state of the art solutions while offering upgrade functionality down to the network level and iii) can reuse existing Contiki network protocols for dynamic updates without requiring modifications to the code.

Introduction

Currently, in the developed world, individuals own a limited number of wireless devices including a mobile phone, a PC, a tablet, camera and a TV. The applications running on most of these devices are able to perform automatic updates to relieve the user from a relatively difficult and time consuming upgrade process. IoT brings a third wave of computing innovation, following the personal computer and mobile phone wave. The Internet of Things (IoT) is typically envisioned as the integration of small, cheap and/or energy efficient wireless radios in everyday objects. Most IoT devices are constrained in terms of processing power and memory storage to keep the unit price low, as well as in terms of energy since they are typically battery powered. This trend will lead to a further increase in i) the number of devices per person and ii) the number of devices per square meter, thereby introducing the need for efficient upgrade ability and reconfigurability of embedded devices, as recognized by Atzori et al. [1], De Poorter et al. [2], Fortuna and Mohorcic [3] and the ETSI RRS WG 2 [4].

In addition, whereas wireless technologies and standards are changing at a rapid pace, embedded IoT devices will have a longer lifetime, typically a decade or more. To cope with such a huge amount of devices, deployed in rapidly changing wireless environments, we envision the need for update mechanisms that not only allow new IoT applications, but also allow to download new or updated network functionality, including MAC implementations, over the air from a repository (similar to an app store). This would, for example, allow to switch between a ZigBee [5] and WirelessHART [6] protocol stack, both of which use the same physical layer.

The most popular wireless sensor network/IoT operating systems, TinyOS [7] and Contiki [8], were not designed to support such dynamic upgradeable environments. In these operating systems, while the applications can be altered using partial code upgrades, altering the network stack can only be achieved using full firmware upgrades. This approach has several major drawbacks. (i) Distributing a new firmware over-the-air consumes a significant amount of energy. (ii) Extending or modifying large network stacks is time consuming and error prone. (iii) The lack of code re-usability slows the overall progress in the area. Other systems, such as SOS [9], allow more modular upgradeability but (i) have not been demonstrated to support network stack upgrades, (ii) require modifications to existing network protocols and (iii) do not use standard upgrade mechanisms and file formats.

In this paper, we claim the following three contributions. The first and most important contribution of the paper is GITAR, a Generic extension for Internet of Things ARchitectures that enables run time updates of network stacks and applications on constrained devices using standard file structures, tools and methods. GITAR supports position independent dynamic linking of components at run-time from the application level down to the MAC layer. The second contribution is a proof of concept implementation that is fully backward compatible with the existing Contiki network protocols, thus extending an already popular IoT solution. The third contribution is the theoretical and experimental evaluation of the reference implementation which demonstrates that GITAR (i) is up to 14% more efficient in terms of memory usage; (ii) has less or the same run-time CPU overhead as the state of the art while offering beyond state-of-the-art network level upgrade functionality; and (iii) has a lesser or equal energy cost for deploying the dynamic components.

The remainder of this paper is organized as follows. First, Section 2 describes and analyses scenarios in which run-time software upgradeability is required. Section 3 gives a brief overview of related work. Next, Section 4 described the architecture of the proposed system that enables all the scenarios identified in Section 2. The reference implementation of the architecture is discussed in Section 5 while Section 6 evaluates the architecture both theoretically and experimentally. Finally, Section 7 concludes the paper.