Geographical dispersal of name servers.

Jim Reid jim at rfc1035.com
Wed Oct 30 17:33:52 UTC 2002


>>>>> "Martyn" == Martyn Hills <martyn.hills at attbi.com> writes:

    Martyn> Is there any real advantage in placing DNS servers at
    Martyn> disparate geographical locations around the world?

Yes. It prevents one location from being a single point of failure.

    Martyn> I have web servers located in the US which are used by
    Martyn> customers located in the Far East and Australia –
    Martyn> Someone suggested that we should site DNS servers in those
    Martyn> geographic regions to speed up the resolution time.

That may be true, though whether anyone can really see that speed up
it doubtful. Remember name servers cache answers. So if your DNS data
has reasonable TTL values -- hours or days -- all you speed up is 1
one lookup out of probably thousands made for that name in the TTL
interval. And even then, that one lookup might take 50ms instead
250ms. Who really cares? Can any end user even tell the difference
between those 2 lookup times?

    Martyn> Perhaps I'm missing something, but I cannot see what would
    Martyn> cause a Internet user located in Australia to be directed
    Martyn> to the name server located in that country in preference
    Martyn> to one located anywhere else.

Name servers generally measure the round trip times to the servers
they query. That allows them to send their lookups to the fastest
responding server most of the time.

Remember too that the Internet does not observe geographic
boundaries. Somebody in Sydney say might find that a name server in
Amsterdam is nearer/faster than one in Melbourne. It all depends on
routing/peering policies and where the cables go.


More information about the bind-users mailing list