next up previous
Next: Implementation and Analysis Up: Scaling Mechanisms for Name-Based Previous: Explicit Aggregation

Redirection

 

Not all hosts with names in a given suffix are connected to the network globally advertising that name suffix. For example, there may be hosts with stanford.edu names scattered throughout the Internet, even though most Stanford names are located together. It does not seem feasible to advertise all of these names globally. Such hosts could simply be assigned fixed addresses by the content router operated by Stanford. However, we see some benefit in allowing local flexibility in address assignment without updating a remote server-- particularly in situations where network address translation is being used, or in mobile networks.

INRP provides a redirection mechanism for finding isolated names not advertised in the the name-based routing system. Such names have records indicating their actual topological location in the Internet in terms of a more well-known name. When a request is answered with a redirection record, the client (or the first-hop content router acting on its behalf) restarts the query using the proper name. For example, if the host gritter.stanford.edu is located at Berkeley, a name lookup might return a redirection to guest32.berkeley.edu. This redirection mechanism trades fate-sharing and name lookup latency for decreased routing state; economic factors may well determine what names appear in routing tables and which are found through redirection. That is, an ISP may charge per name it places in the routing table, so an organization weighs the cost-benefit of having a name handled by the ISP versus incurring the redirection cost on a name.

This secondary mechanism is really only needed when using NBRP to replace all DNS usage; for content routing, name-based routing tables contain only site and content volume names rather than host and network interface names.



Mark Geoffrey Gritter
Fri Jan 19 09:19:43 PST 2001