Geometric information routing

In response to the increasing traffic volume in the Internet for applications such as (mobile) video and cloud computing, various proprietary technologies enabling content distribution have been developed that rely on caching and replication. Being deployed in silos, it is not possible to uniquely and securely identify named information independently of the distribution channel; moreover, these different content distribution technologies are typically implemented as an overlay, leading to unnecessary inefficiency. By introducing uniquely named data and name-based data access, Information-Centric Networking (ICN) enables data to become independent from their network location, application, storage support but also means of content exchanges enabling in turn in-network caching and replication. However, content name spaces have not been designed to sustain forwarding performance and forwarders scaling contrary to IP addresses which can be efficiently aggregated, summarized and translated. Consequently, alternatives such as name-based routing, which aim at better accommodating information/ content routing in the Internet, would also become the scaling and performance bottleneck. To address these problems, this paper proposes a third alternative: geometric information routing on universal content locators. This technique operates by associating to content identifiers (names) a content locator taken out of a geometric coordinate space from which a routing path (more precisely, a geodesic) can be derived without requiring non-local information. Upon querying specific content multiple locators can be received enabling the receiver to select the (geometrically) closest locator. Since it is based on local information only, routing on such locator space is less memory consuming than non-local information dependent routing. We analyze the performance (in terms of memory space required to locally store routing states and the resulting routing path stretch) and compare them against path-vector routing.