Lilith defines a new interconnection architecture
An interconnection architecture for spontaneous networks based on label switching #
Communication between devices of every kind keeps becoming more common, giving equipment that used to stand alone new capabilities, at home as much as at work. Many technologies — wireless ones in particular — have distinct characteristics suited to their own application domains. The resulting network is therefore heterogeneous, and it changes often as equipment is added and removed, largely because devices move along with their users.
Such a network is typically spontaneous: devices join and leave over time with no predictability. Internet access can improve how they work — clocks synchronise themselves against a server, a television can show schedules and reviews, an oven can display a recipe. This spontaneous network of devices sits at the edge of the Internet. The usual interconnection architecture, the Internet’s own, is a poor fit for it, because it does not account for what makes spontaneous edge networks different. In two studies, published at the Workshop on Future Directions in Network Architecture 2004 and at the MobiQuitous 2004 conference, we present an interconnection architecture suited to spontaneous edge networks.
An interconnection architecture defines how different devices communicate over a shared network. On the Internet, that architecture rests on IP: data is split into packets, each forwarded independently from source to destination. Two consecutive packets may well take different paths and arrive out of order. The Internet’s success has proved the architecture effective.
Spontaneous edge networks are a different case, and they call for another approach. First, their spontaneous nature leads us to treat them as ad hoc networks, working without infrastructure such as access points or central servers. To cover areas of any size while still allowing two distant stations to communicate, ad hoc networks are multi-hop. Every device in a building can then communicate without any dedicated infrastructure being installed.
Second, the spontaneity of these networks makes every path between two devices temporary. One has to account for, say, a personal assistant carried by someone walking. When a path is about to disappear, that has to be detected quickly and another path used. Having backup paths also limits the effect on user-visible traffic of topology changes, which can happen abruptly.
Finally, the path packets take should be determined per type of communication between source and destination: video-conferencing and weather statistics do not share the same constraints, particularly in latency between sending and receiving. Classifying traffic at the level of the interconnection architecture makes it possible to give each class a priority, or a minimum guaranteed throughput.
These constraints led us to design Lilith, an interconnection architecture for spontaneous edge networks. Lilith is based on MPLS, a layer 2.5 sitting between the link and network layers, and it is MPLS rather than IP that forwards the data. Using MPLS answers the constraints above elegantly, without changing the operating system or the applications running on it. A prototype was developed for Linux and let us compare Lilith’s performance against IP forwarding. The degradation is slight — around 0.3% in throughput — and has to be weighed against what the architecture provides.
- [MobiQuitous 2004] Vincent Untz, Martin Heusse, Franck Rousseau and Andrzej Duda. Lilith: an Interconnection Architecture Based on Label Switching for Spontaneous Edge Networks. In Proceedings of MobiQuitous 2004, Boston, USA, 22–26 August 2004.
- [FDNA 2004] Vincent Untz, Martin Heusse, Franck Rousseau and Andrzej Duda. On Demand Label Switching for Spontaneous Edge Networks. In Proceedings of SIGCOMM Workshop 2004, Portland, USA, 30 August – 3 September 2004.