there has never been a shortage of volunteers to run root servers. i'm not sure that the exchange points are good spots, since the folks who run exchange points (MFS, NASA, Pac Bell, etc) usually know a lot more about the link level than the network level -- and besides, i'm not sure that a root name server ought to have an ASN and run BGP4, which means that it will have to be behind a router that _does_ have an ASN and run BGP4. if this is to be the case, i'd rather see the routers inside NSP offices, where UNIX experts and network experts are more plentiful than at the exchange points.
Actually, I think that it might be interesting to look at the possibility of putting Root DNS on the Route-Servers.
to that end, NS.ISC.ORG is one 10Mb/s hop away from BADnet (barrnet-alternet- digital) in the DECWRL computer room. it's likely that i will shortly add an SMDS T1 connection to the CIX cloud for other reasons, and that will help a tiny bit (only a tiny bit, since Alternet has T3 to that cloud from the router i peer with in that room.)
work is underway (by the Postel-Mockapetris-Vixie-Kosters quadumvirite) to implement Bill Manning's suggestion of putting all the root servers under a single domain, which will let DNS' name compression start winning for us. if this works out, we should be able to just about double the number of root name servers. NSP's with multiple T3's to geographically disparite exchange points will be given strong preference. sites outside the United States will be given strong preference.
Interesting... That should definitely help.
the root servers are not currently suffering from load (my own server does about 100 queries per second, which is about 10% of the capacity of my little 66Mhz/64MB BSD/OS machine). the goal here is to reduce load on the wide area net rather than on the root servers themselves. and also to increase the likelihood that any given host can reach a root server during times of wide- scale connectivity problems (which seem to be more common lately?)
True. However, the nice thing about a RNS on an XP is that the RNS is one hop from EVERY provider at the XP, and is independent of OTHER providers. I agree that the XP management shouldn't manage the RNS. However, I would be QUITE comfortable with MERIT running it on the RS or adjacent machine.
For security and stability reasons (aswell as political) they should not be run by a single organisation.
they never have been.
Although I agree that ALL RNS's shouldn't be run by a single organization, I would not have a problem with the RNS's at the XP's being run by the RADB group. Owen
Although I agree that ALL RNS's shouldn't be run by a single organization, I would not have a problem with the RNS's at the XP's being run by the RADB group.
I would. Elise and her team are experts at what they do, but running a root name server is not quite the same thing.
Actually, I think that it might be interesting to look at the possibility of putting Root DNS on the Route-Servers. Why? True. However, the nice thing about a RNS on an XP is that the RNS is one hop from EVERY provider at the XP, and is independent of OTHER providers. I agree that the XP management shouldn't manage the RNS. However, I would be QUITE comfortable with MERIT running it on the RS or adjacent machine. <I'm about to projectile vomit all over my keyboard> I don't quite know how to do this without insulting the nice folks at Merit (sorry Elise), but I can't see any possible rationalization for throwing tax dollars in that particular direction for this sort of service. We've just spent a lot of time and effort making it possible to completely remove central "authorities", like Merit, from the operation of the net. The next target towards a decentralized network is the InterNIC followed by the RNSs. That bizarre idea about alternative name hierarchies is actually an extremely attractive suggestion, although I had a bit of a problem keeping down my dinner when I first read it. Although I agree that ALL RNSs shouldn't be run by a single organization, I would not have a problem with the RNSs at the XP's being run by the RADB group. The expertise required for DNS is an almost totally disjoint subset of the expertise required to handle name service. There's really no reason whatsoever to toss these together. In fact, I find the entire idea of putting general purpose machines at the exchange points to be questionable. The whole point of having an "Internet" is so that we can utilize services throughout the Internet. If we have to start placing RNSs at exchange points, we've seriously lost sight of that goal. As long as there are -enough- of them, and they are reasonable well distributed, there is no value add, and in fact, there is a considerable value subtract.
participants (4)
-
bmanning@ISI.EDU
-
owen@DeLong.SJ.CA.US
-
Paul A Vixie
-
Paul Traina