On Mon, 01 Sep 2003 06:25, you wrote: > On Mon, Sep 01, 2003 at 05:51:19PM +0000, Shane Hollis wrote: > > I think we are beginning to go around in circles here .... > > Probably... :) > > > When the address is found, the > > four levels of DNS server that it has passed through will cache the > > result. > > Important to note here, that they wont actually cache anything, since > they're authoritative for what they were asked for. > > e.g., the root server was asked 'where's .nz' ? it answers. > the .nz server was asked 'where's .geek.nz' ? it answers. > > etc. The *requesting* server, on the other hand, will cache these results, > with the TTL's given, so that it wont ask a root server again for .nz, > until it expires.. which is.. (goes to look...) > > [EMAIL PROTECTED]:~> host -vvv -t NS nz > ;; QUESTION SECTION: > ;nz. IN NS > > ;; ANSWER SECTION: > nz. 86378 IN NS ns5.dns.net.nz. > ... > > looks suspiciously like 86400 secs, or 1 day. > > etc etc. > > > We are also agreed you do not always go to the source to find an address, > > rather you can get it from an 'upstream' trusted DNs servers cache. > > Upstream from you. there is no law stating your DNS server must be a client > of another. you can set up your DNS server to not 'forward' unknown's, but > to recursively find them out itself, which in this case, means you *will* > get the result from the source. (and then cache it) > > > Having all agreed on that ( I hope) that brings us back to the thing > > someone originally picked me up on.... if it is faster to disperse IP > > changes from a backbone or the small pimple on the butt of the internet > > DNS server. > > Ok, you're starting to make sense... and yes, this was the point I was > tripping you on.. (well, trying.) > > > It is still my contention that to make an internet wide change of IP > > address, one that will eventually affect all caches and all DNS's then it > > is better to do it from a central location with lots of traffic ( > > especially DNS traffic) if you want to get the fastest possible way of > > spreading the changes. This however does not apply to new addresses. It > > also doesn't mean a change will not be spread from the pimple of the > > internet, just that it will take longer. This length of time would be > > especially important to someone using one of those dynamic DNs lookups > > such as zoneedit. If your ip chages, you want that change spread as fast > > as possible so there is as little ooops where is that machine now as > > possible. > > *NO*. Time to propogate is IRRELEVANT when you're talking about the server, > or location. Not quite ... if one person at a university of 10 000 people asks for a change then the Uni's DNs gets the IP and all people at the university have instant access to that IP. Yes?
That negates incorrect caching and trouble for all 10 000 people. A big server in the states (Server A) regulalry gets hit by many other servers (Server B - Z) who serve lots of people. If the big server(A) is correct then it will give all those other servers (B-Z) correct info. They in turn only need one person to ask for the new IP address. This means that one person from each of those servers(B-Z) only needs ask and the correct info is available to all the people. The size of the server doesn't determine the look up speed but it does determine the number of users who now have access to the correct information. In an ideal world anybody who does a lookup will get to the correct site every time. In the real world there are stuff ups with caching and it doesn't happen. Having your IP registered in a big central server does several things. 1: - Quick return time for lokups as you are working on the backbone of the internet not a small byway. 2: - Less steps to return the lookup as you are closer to the places many people will use and so your cache or info is more up to date. (ie if most of my work is for say an american company then having the record in the USA means the servers that most need to know the info are closer to their home so less steps ( ie don't have to go to NZ and then a local ISP to find it). 3: - More people able to access the same lookup as there are more people using the same DNs server to look up the same address. Therefore less traffic on the net. To me it makes sense that if my domain is primariy used by USA people then register your domain record in the states. If it is used mostly by Taiwanese, register it in Taiwan. We are only talking a small matter of degree but I know trying to find some of those esoterric chinese hacker addresses has been a pain because it sometimes takes multiple goes before I can get a return on a dig or host lookup owing to the speed and stability of some of their networks. Anyway, my head hurts a lot, I am going to bed. cheers, shane > > Propogation time is ONLY relative to the TTL for the record. doesnt matter > if it's a pimple, or a huge whopping great puss ball. if the pimple is > authoritative for a domain, it will still take the same amount of time for > the change to 'propogate' as it would from the puss ball. (as long as the > TTL's are the same). > > Now.. here's a point for you: 'propogation' is a really bad word to use for > this purpose, as what is really meant, is *expiry* of all old records > cached aroung the net. Nothing actually get's propogated anywhere, except > to the secondary servers for the domain, via AXFR :) > > > does this sum it all up? > > Sort of... > > Mike. -- Shane Hollis Notes Unlimited New Zealand Ph: 021 465 547 Email: [EMAIL PROTECTED]
