A partitioned network has several subdomains (1, 2, 3,) each having a respective access server (10, 20, 30) and each serving a plurality of end users (11, 12; 21, 22; 31, 32). Each access server (10, 20, 30) is connected to a respective content cache (13, 23) or a set of such caches (130, 131, 132), which stores content and downloads it to end users on request. An alternative routing (19, 29) may be available between an access server (10) and a content cache (23) other than its associated cache (13), for use in exceptional circumstances. All the content caches (13, 23), are given the same IP address, “W.X.Y.Z”. Each access server (10, 20, 30) operates exclusively in a geographic sub-domain, and recognizes the IP address W.X.Y.Z as relating uniquely to its respective associated content cache 13, (23, 23) This means that the configuration of each access server (10, 20, 30) to handle content caching/distribution/streaming can be the same regardless of its location in the network. Any end user (10) can therefore obtain content from the cache (13) which is topologically closest, using the common IP address W.X.Y.Z. No address translation or mapping is required to route requests to the correct local content server. If the access server (10) is unable to connect to the associated content server (13), it is configured to forward traffic to another content cache (23), by tunnelling the request to an access server (20) associated with that second cache. The second access server (20) recognizes the address W.X.Y.Z as being that of its own associated content server (23), rather than the content server (13) associated with the first access server (10), and forwards the request accordingly. The content can then be returned to the address of the requesting user.