Okay, let me say this more precisely. I've seen too many occasions over the years where DNS was broken so badly that the only way to get things fixed, or to get work done, or to keep applications operating, was to bypass DNS - either by typing in an IP address somewhere. And of course those alternative mechanisms also break - but the combination of mechanisms work better than DNS alone. And of course one of the reasons that the combination works better is that the addresses of "important" hosts rarely change. I'm not saying that DNS can't be improved to be more reliable - clearly it can. I'm not saying that DNS hasn't been improved - clearly it has, though old habits die hard and users are slow to change what "works" for them. But putting _all_ addresses in DNS makes DNS failures a lot more critical than they are now, and there are good reasons for being reluctant to do that. >> And really, there's no way I'd trust DNS to do this. I've spent too >> many years watching it break. --Keith >> > > i suspect that you're measuring the wrong thing, or that you're not paying > attention to the "what" that you're measuring. in a every distributed system > of sufficient size, there is always something broken somewhere. the sysadmins > at ISC were for example concerned when the trend of broken f-root hosts got to > the 1-a-day level until someone pointed out that once you've got more than 100 > systems at least one will always have something wrong with it and it's a good > thing we put two in every POP and have a lot of POPs isn't it? > > yes, DNS is always broken. so is the routing table. so is the airline system > and most road systems and the stock market. and it always will be broken, > since in systems of sufficient size, entropy and human error are signigicant > enough to be noticed. if you don't want to use something that will break, you > ought to start by pulling the power cord out of all your servers and routers. > > it's just not reasonable to demand 100% uptime from a million-node distributed > system where most of the nodes are operated by other people. doesn't matter > if the nodes are BGP routers, web servers, DNS servers, or botted home PC's. > > odell's 8+8 relied on DNS for location->routing mapping and that could be one > of the reasons it had so little support. but in the decade+ since then, DNS > has scaled better than the routing system. odell had a reasonable design but > it lacked the architectural purity of... whatever it is we're using instead. > > _______________________________________________ > Ietf mailing list > Ietf@xxxxxxxx > https://www1.ietf.org/mailman/listinfo/ietf > _______________________________________________ Ietf@xxxxxxxx https://www1.ietf.org/mailman/listinfo/ietf