[ntp:questions] More Granularity in the US in the NTP Pool

Ryan Malayter malayter at gmail.com
Sun Sep 9 12:25:40 UTC 2007


On Sep 9, 4:21 am, Uwe Klein <uwe_klein_habertw... at t-online.de> wrote:
> The next best thing imho would be to select by lowest hopcount and ignore
> any geographic details.

You're asking for a mapping of the Internet into a virtual coordinate
system based on hop count or RTT. There has been a lot of research in
that area (check Google), but unfortunately, it is very resource-
intensive. Basically, every participating note has to periodically run
traceroutes to clients and share state information with the DNS
servers. This is what people pay Akamai big bucks to do transparently.

We're already using OASIS data in Guillaume's geographic DNS beta,
which does something similar. But to make it work as you describe,
we'd have to have each participating server run additional OASIS
software, as well as commit a lot more bandwidth for network
measurement. See the papers at:
http://oasis.coralcdn.org/

All of these virtual coordinate systems assume the client is
"oblivious" to network locality, such as WWW clients. Fortunately, NTP
is *not* oblivious to network conditions. All we really need to do is
return a subset of reasonably close servers, and NTPd will follow the
best one. This is why Guillaume has focused on very granular (1 degree
of latitude/longitude) geographic location as a starting point.

Regards,
Ryan




More information about the questions mailing list