designing the DNS from the scratch

[Matthew Seaman]

On 2017/07/10 14:16, Matus UHLAR - fantomas wrote:
>>> But you do know the approximate speed of light in a vacuum?
> 
> there's always dark in my vacuum, so the speed of light doesn't apply
> there.
> 
> On 10.07.17 09:02, wbrown at e1b.org wrote:
>> More importantly, what is the speed of light in a fiberoptic connection?
>> Speed of electrons in copper wire?
> 
> speed of electrical field, which is the same as speed of light.
> electrons are much slower.
> 
> however, the longest distances on earth are about 20000km, which requires
> at least 67ms for signal to get there and 133ms to get back.
> in reality there's some small delay on each network device in the path, so
> the 3ms can only be achieved on short distances.
> 

Indeed.  Assuming the OP was talking about providing an authoritative
service -- that is, to allow the rest of the world to look up their
customer's domains -- then if they went back to their customer with a
more realistic target of say a 95th-percentile limit of a sub-50ms RTT
for users in urban North America, Europe, Russia, Japan and other
locations with a well developed Internet infrastructure, that could be
achieved by putting DNS servers in strategically located POPs on each
continent and using anycast routing to direct traffic to the nearest
location.

Which would be eye-wateringly expensive to do for just one client,
unless they needed about as much capacity as a middle-sized ccTLD.

Or you could buy a service from one of a number of DNS service providers
who provide pretty much exactly what I described.  That will still be
quite expensive, but not to the extent that it would cause inadvertent
emission of bodily fluids.

On the other hand, if they were talking about providing a recursive DNS
caching service to allow their customer's servers to look stuff up from
the internet, then a 3ms RTT is not impossible so long as

   * the DNS machines are sufficiently close to the client's machines
     that you can readily achieve sub-3ms ping RTTs between them

   * the 3ms limit *only* applies to responses from cached data.

There's clearly no way you can guarantee <3ms if your recursive server
needs to talk to a machine on the other side of the planet where it
takes at least 200ms just to get packets there and back again.

	Cheers,

	Matthew


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 972 bytes
Desc: OpenPGP digital signature
URL: <https://lists.isc.org/pipermail/bind-users/attachments/20170710/ddd74332/attachment.bin>