On the efficient segmentation of network links

The Internet is experiencing a change in paradigm: once determined by text-based applications, currently, various types of audio-visual content already adopt the major part of the bandwidth. Audio-visual applications and multimedia data impose individual demands on reliability and delivery time, which are not sufficiently covered by available transport protocols. In addition, a more efficient use of network resources such as link bandwidth or the spatial distribution of network routes is essential. Therefore, a split of a given transmission path into several independent areas, where each contains a set of similar, adjacent network links is appropriate. An individual transmission and error correction scheme can be applied to such a group of related links. In previous work it has been shown that there is an optimum number of segments. Too few or too many limit the achievable efficiency, i.e. increase the required redundancy information. The search for this optimum, however, is a complex task not applicable to real-time adaptivity. This paper proposes a decentralized, metric based mechanisms to decide where a network path should be intersected, leading to a lower complexity for the retrieval of these areas. We also investigate an example where the proposed clustering approach leads to an overall redundancy information of about 1/4 of the redundancy information obtained by a traditional end-to-end approach and thus is very close to the Shannon limit.