[not sure this got out the first time; please excuse any duplicates.] Bill, In essence I disagree with you. We are building dependencies on new parts of the infrastructure, but we started doing this the moment we introduced a router, continued it with DNS, and continue it in new and different ways. We're also doing things with the network that we weren't doing a few years ago, so some of the costs are well worth the enterprise's expense. Basic DHCP makes laptop use practical. 7x24 remains a goal. Thus a draft on how to do DHCP fallback. Thus the unending arguments about so-called "multi-master" DNS. Of course it remains the responsibility of the application developer to understand how to provide a reliable service in the face of an application failure. Also, the developer needs to make realistic assumptions about how reliable the lower layers will be. This doesn't relieve the Internet Area or any other part of the IETF from considering 7x24, especially if the goal is 99.999% uptime (that's 5 minutes a year). Routing convergence is something that has and remains a big issue on large public networks. Oddly, with the exception of the end nodes, LANs are typically in better shape. There is perhaps a good BCP or FYI to be found here. Bill Manning wrote: > We are building networks that have required infrastructure support > before the end nodes can be usefully deployed, e.g. CA/Key servers, > DHCP/DynamicDNS, policy servers, etc. > > Its not that we can't build them, its that it is a change in > a fundamental IP precept. Just a thought. > > > > --bill -- Eliot Lear [EMAIL PROTECTED]
