IEICE Workshop on Active Network Technology and Applications (ANTA) ©2002, Tokyo, March 2002, pp. 71–74.
| Bernhard Plattner | James P.G. Sterbenz |
| Computer Networks and Engineering Laboratory | Internetwork Research |
| ETH | BBN Technologies |
| Zürich, Switzerland | Cambridge, MA, USA |
| plattner@tik.ee.ethz.ch | jpgs@m.ieice.org |
+1 41 1 632 7000 |
+1 508 944 3067 |
Active network research has made significant strides over the last several years. While current research focuses on using active networking technology in wired networks, with emphasis gradually moving from active language environments toward dynamic service deployment, highly mobile and ad hoc networks open a completely new and challenging area of applications. We argue that active network technology will be most instrumental in situations with a relatively high ratio of the availability of processing and memory to bandwidth, which applies to many mobile computing situations. Furthermore, ad hoc nodes already have the characteristic of being both end and intermediate system, and thus share many of the resource and security issues with active nodes in general. There is large range of algorithms and protocols to which active technology can be applied (and benefit from), including autoconfiguration, self-organisation, resource discovery, novel application and content-oriented routing, simultaneous protocol and addressing mode support, dynamic service deployment, and session control using secure overlays. Unlike the traditional Internet, ad hoc and sensor networks do not have significant agreement on protocol standards, and thus there is a unique opportunity for active networking to actually facilitate the deployment of mobile wireless networks by providing a systematic and flexible protocol framework, without requiring overstandardisation of individual protocols.
Mobile wireless active network, adaptive network, resource tradeoffs multimodal link MAC protocols, magnetic routing, rapid service deployment.
Mobile wireless networks are ripe for the application of active networking technology. The dynamic behaviour of mobile nodes and the inherently complex and time varying characteristics of wireless channels provide significant additional motivation for intelligence in network nodes over wireless networks. Indeed, much of the research in the MANET (mobile ad hoc network) community is struggling with how to do this; active networking provides a systematic framework that requires less a priori agreement on algorithms and mechanisms leading to standards. There may be real opportunity here to provide the infrastructure that enables ad hoc networking and its interworking with the traditional Internet before standards are codified.
There are a number of common assumptions about the Internet that are likely to be invalid in the future; some of these have particular importance in the mobile wireless networking environment.
Active processing is enabled by the decreasing cost of processing and memory, but it is actually the ratio of the availability of processing P and memory M to bandwidth B that is important for a large class of active applications [1]. In particular, it is this ratio that determines how much active processing can occur per flow or transaction. While processing and memory are more abundant, so is bandwidth in the wired network. High-speed links (e.g. DWDM OC- 192 or OC-768) have an inherently low P+M/B ratio.
Wireless links have significantly more modest bandwidth capacity, however. To the degree that it is channel limited, there is opportunity for significant active processing. To the degree that the node itself is power limited, a more delicate balance exists between processing and communication transmission/reception. This is particularly the case for sensor (sub)networks and sensors that are part of a personal area network.
Note however, that active processing may help improve transmission and reception efficiency, by controlling transmission power and limiting the time that transmitters and receivers are powered. A final consideration in this area is that the same limited bandwidth in wireless networks also limits the ability for mobile code as a mechanism to transport active code. While occasional protocol and active code provisioning is likely to be feasible (exploiting the broadcast medium), highly dynamic code provisioning must be traded against application goodput.
One of the fundamental differences between an active node and a conventional router or switch is that an active node has characteristics of both an end system and intermediate system. The active node reference architecture and security architecture recognises the need to control resources so that active applications do not interfere with one another and the normal forwarding service of the node. Commercial implementations of active nodes will need to have all of the filtering and classification functions in the hardware fast path of each interface, and per flow and datapath active processing located with each link interface.
Mobile wireless ad hoc nodes inherently serve both as end systems and as intermediate systems that pass transit traffic to allow the construction of a multihop network. Thus, many of the resource and security issues are common to both ad hoc (passive) networks and active networks. Generalising the active networking solutions to the needs of ad hoc networks is critical. There are a number of characteristics of mobile wireless networks that make these problems even more difficult so solve, such as limited bandwidth, unreliability of links, and the likelihood that infrastructure facilities such as certificate and key servers may be inaccessible.
We can think about some of the implications on the active node reference architecture. In the current model, all activity occurs at or above the network layer. In a mobile wireless active network, additional control is needed. On the receive interface, physical layer coding may have to be dynamically sensed, and channel conditions monitored to provide input for active processing. Active code may schedule and control when the receiver is powered on and able to receive. Frames may be filtered for active processing based on link and MAC layer headers. In the control plane, link and MAC protocols and parameters may be dynamically changed, or simultaneously supported. On the transmit side, active processing may control link and MAC layer framing, transmission parameters and power-on activity, and physical layer coding.
Note that this processing on the interface drives toward a distributed architecture in which active applications operate on the interfaces, rather than in central processor of the node. This is a requirement for any high performance node with multiple interfaces, which already distribute classification, address lookup, and output scheduling in interface modules.
Sensor networks and computational fabrics (e.g. electronic textiles) present particular challenges and opportunities. The challenges come from the extreme simplicity required in the node resources (processing, memory, and energy consumption). The opportunities come in the ability to use active code to achieve an overall reduction in these resources, and to embed evolving code into sensors to dynamically adapt to changing channel conditions and application requirements.
There are a large number of opportunities for active processing in mobile wireless networks. Note that there is a fine line between calling some this “active” versus an algorithm built in the node. Doing it as active processing allows dynamic provision without codification of standards and preconfiguration, as well as natural coexistence of multiple algorithms. Some of these protocols and algorithms include:
All of these protocols can benefit from context-awareness. For example, many mobile services can exploit geolocation dependence, and high-velocity mobile nodes can use trajectory information to optimize routing topology and cluster federation updates. Similarly, the ability to sense administrative location can assist in the optimisation of the relationship of layer 3 cluster topology to a secure overlay structure.
Active networking allows the support of arbitrary protocol processing. At the network layer, this can range from custom protocol processing (e.g. multicast, congestion control), through simultaneous protocol version support (e.g. IPv4 and IPv6), to support for completely distinct network layers and forwarding modes (e.g. IP forwarding, characteristics -based, and geolocation such as Cartesian or diffusion; see [2] for more on this). Conventional wired networks have relatively fixed topologies and link technologies, however.
Wireless networks provide the opportunity to move active processing down through the MAC, link, and physical layers. Active processing can be used to dynamically adapt parameters for MAC (e.g. TDMA slot length and CDMA codes), link (e.g. FEC strength and frame length), and physical transceivers (e.g. channel coding and transmission power). At the extreme end of the spectrum, the entire protocol suite can be adapted to the dynamic nature of mobile nodes and changing channelcharacteristics. For example, nodes may dynamically negotiate with one another to select the appropriate MAC protocol (e.g. TDMA vs. CDMA) and spectrum use to adapt to mobility, fading channels, andtraffic conditions. Software radios and programmable hardware naturally enable this technology. Optimisations such as ILP (integrated layer processing) can extend all the way down to the physical layer when appropriate, in an arbitrary mix of embedded controller software and programmable hardware.
Magnetic routing is a paradigm in which messages and information are routed based on a field that governs the attraction of packets toward a destination. We have considered two flavors of this:
Service creation and deployment is challenging enough in the context of wired networks; it is a significantly harder problem in mobile wireless networks due to:
Mobile nodes can adapt to these conditions by preparing to assume some of these service infrastructure roles while the primary resources are unavailable. For example, nodes can cache name resolutions and address bindings to temporarily perform binding and routing when these servers are unavailable. Inter-node active processing can allow nodes to assume these functions for one another without requiring all of them to contain the processing and memory to perform all of them. Clearly, this load balancing should also be power aware.
Many of these ideas were developed with input from and discussions with: Alden Jackson, Rajesh Krishnan, Craig Partridge, Ram Ramanathan, and Martha Steenstrup at BBN; Polly Huang and Burkhard Stiller at ETH. This work was performed at ETH under the FAIN program and at BBN Technologies funded by DARPA (Doug Maughan PM),under contract F30602-99 C-0131 issued by Rome AFRL
IEICE Workshop on Active Network Technology and Applications (ANTA) ©2002, Tokyo, March 2002, pp. 71–74.
Last updated 18 November 2003 –
Valid XHTML 1.1 –
Lynx inspected –
W3C AAA Conformance
©2003 James P.G. Sterbenz
<jpgs@sterbenz.org>