There is an increasing number of Internet applications that attempt to optimize their network communication by considering the network distance across which data is transferred. Such applications range from replication management to mobile agent applications. One major problem of these applications is to efficiently acquire distance information for large computer networks. This paper presents an approach to creating a global view on the Internet, a so-called network distance map, which realizes a hierarchical decomposition of the network into regions and which allows us to estimate the network distance between any two hosts. This view is not only a single snapshot but is dynamically adapted to the continuously changing network conditions. The main idea is to use a certain set of hosts for performing distance measurements and to use the so-gained information for estimating the distance between arbitrary hosts. A hierarchical clustering provides the notion of regions and allows us to coordinate the measurements in such a way that the resulting network load is minimized. An experimental evaluation on the basis of 119 globally distributed measurement servers shows that already a small number of measurement servers allows us to construct fairly accurate distance maps at low cost.
展开▼
