[time-nuts] time.windows.com statistics

[Chris Albertson]

On Mon, Jul 14, 2014 at 3:38 AM, Tim Shoppa <tshoppa at gmail.com> wrote:

> Wonder out loud if using NTP server in a load-distributing cloud will just
> intrinsically be randomly cruddy, or if this is somehow engineering a
> source which is often good enough for SNTP users but obviously
> inappropriate for NTP to prevent extra load from non-windows users.

Pool servers seem to work well.  But they are randomly assigned only
once.  Once a connection is made to a pool server it remains connected
to the same server "for the duration".   Best practice is to use about
five pool servers and let NTP sort out which are "best".  Even if
using a GPS you should still configure about five pool servers.

A load distributed load cloud where each data packet goes to a
different server just plan would not work for NTP.  If you look at the
clock selection algorithm the servers would all be rejected and NTP
would likely fall back to the PC's internal clock as the best source.
 NTP's design depends on network path between any two NTP servers
remaining mostly constant or at least changing slowly.  Load sharing
would kill that assumption.

-- 

Chris Albertson
Redondo Beach, California