Energetic performance of service-oriented multi-radio networks
Proceedings of the 6th international workshop on Software and performance - WOSP '07
Wireless devices now hold multiple radio interfaces, allowing to switch from one network to another according to required connectivity and related quality. Still, the selection of the best radio interface for a specific connection is under the responsibility of the end-user in most cases. Integrated multi-radio network management so as to improve the overall performance of the network(s) has led to a number of research efforts over the last few years. However, several challenges remain due to
... e inherent complexity of the problem. This paper specifically concentrates on the comprehensive analysis of energy-efficient multi-radio networking for pervasive computing. Building upon the service oriented architectural style, we consider pervasive networks of services, which are deployed on the various networked nodes. The issue is then to optimize the energetic performance of the pervasive network through careful selection of the radio link over which service access should be realized for each such access. This leads us to examine first the energetic performance of service access for most common wireless interfaces in use today (Bluetooth, WiFi and GPRS) and then introduce a formal model of service-oriented multi-radio networks. The proposed model enables characterizing the optimal network configuration in terms of energetic performance, which is shown to be a NP-hard problem and thus requires adequate approximation. Introduction As in particular targeted by Beyond 3G (B3G) networks, the recent evolution of mobile networks introduces the convergence of wireless technologies, where several radio interfaces may be used concurrently. B3Gcapable devices then hold several radio interfaces, and allow switching from one radio interface to another (e.g., upon disconnection, to save money, to save energy, ...). Still, taking benefit of such a rich networking environment is raising tremendous challenges. Indeed, while operating systems start embedding support for integrated management of multi-radio networks (e.g., Windows Mobile 5 deals with transparent switching from GPRS to WiFi during Internet access upon WiFi connectivity), the end-user is in charge of explicitly choosing and possibly switching networks in most cases. A key challenge for integrated management of multi-radio networks is then to effectively contribute to improving the performance of the networking environment. Multiple performance criteria must further be taken into account. These in particular include network throughput and energy consumption, which are antagonistic performance factors. Additionally, multi-radio network management shall be made as far as possible transparent to the end-users and possibly to application developers, simply requiring them to abstractly specify base connection profile/constraints. The above concern has led researchers to investigate multi-radio network management below the application layer. The connectivity management middleware presented in  provides a connection manager from which the application may get accurate knowledge about connectivity through the various links and related performance, and then decide upon the specific network connection(s) to establish. Network connections may also be adapted by the middleware according to application-defined policy and changes in the networking environment. Adaptive network connection may alternatively be realized in the lower network layer , leading to fully transparent solutions towards optimizing multi-radio networking on the end-user devices. One approach is then to favor the use of the low power radio as long as related bandwidth meets system requirements (e.g., to exchange control messages and even data messages for which the bandwidth requirement is low  ). Coordinated usage of the various radio networks may also be customized for specific networking functions, like resource discovery [17, 7] . Independently of the system layer in which multi-radio network management is implemented, most solutions we are aware of concentrate on optimizing network performance locally, i.e., the decision of which network channel to use for a specific connection is based on local information about network quality. As such, it cannot be guaranteed that this will lead to a globally optimal network usage in terms of performance. However, it seems to be the only tractable way to multi-radio network management . Still, network connection on the terminal should not only be established according to local network quality and bandwidth requirement for the specific connection. Performance of network usage in terms of energy consumption depends on the set of local connections already open due to the non-negligible base energy cost associated with network interfaces  . Obviously, the energetic performance of the multi-radio network evolves as network connections are established and closed on the terminal, possibly requiring adaptation over time. Also this does not solely concern the energetic performance on the given terminal but also the energetic performance of the wireless nodes in communication range. Furthermore, networking capabilities vary among nodes, as it cannot be assumed that all networked nodes embed the very same set of network interfaces and even if they do, they must agree on the networking mode and possibly network channel  . As a result, energy-efficient multi-radio network management is still in its infancy, with several issues to be solved before it can be effectively deployed. This paper specifically concentrates on the comprehensive analysis of the energetic performance of multiradio networking in the context of pervasive computing. Indeed, multi-radio networking is a key enabler of the pervasive computing vision, as it promotes anytime, anywhere access to the digital world, whether proximity-based or not. Improving energetic performance of the multi-radio network will allow enhancing network connectivity, while optimizing autonomy of the wireless nodes. The approach to pervasive computing that we promote lies in service orientation with autonomous pervasive services being deployed on the networked nodes, whether mobile or stationary, wired or wireless, resource-constrained or resource-rich  . Pervasive services abstract the various digital resources, from base sensor/actuator (e.g., display) to advanced applications (e.g., nomadic collaborative gaming). Then, services network together according to required and provided functionalities, composing their functions to realize rich distributed services [13, 14] . The actual networking of services depends on the technologies embedded on the hosts of the services' clients and providers since access to a service requires the service's client and provider to use a common network interface to communicate, a common service discovery protocol to find each other, and a common interaction protocol to understand each other (Section 2). Further, energy-efficient networking of services shall account for the respective energetic performance of the underlying radio interfaces (Section 3). This leads us to introduce a formal model of service-oriented multi-radio networks, from which to derive networking configurations that optimize the networks' energetic performance (Section 4). As shown, the problem is NP-hard and optimal energetic performance can only be approximated. As part of the PLASTIC project , we are currently developing a middleware for pervasive service-oriented computing, embedding support for energy-efficient multi-radio networking (Section 5).