[ntp:questions] General ntp architecture question
unruh at wormhole.physics.ubc.ca
Mon Aug 16 17:05:55 UTC 2010
On 2010-08-16, Richard B. Gilbert <rgilbert88 at comcast.net> wrote:
> unruh wrote:
>> On 2010-08-16, ask <ask at develooper.com> wrote:
>>> On Aug 2, 1:35?am, konsu <konrad.azzopa... at gmail.com> wrote:
>>>> Hello list,
>>>> I work for an investment bank with 300 UNIX servers, around 3000
>>>> workstation PCs and would like to ask some questions to more
>>>> experience users.
>>>> a) Are there any banks relying on ntp pool project or should we
>>>> consider having our own GPS clock ?
>>> If you have any sort of legal and practical requirement for keeping
>>> accurate and tracable time you need your own "stratum 1" source.
>>> I'd get a GPS timeserver for each physical location (unless you trust
>>> your internal network very much) and then on each server and
>>> workstation, as others suggested, you configure to use your various
>>> GPS-based ntp servers.
>>> Meinberg donated a couple of GPS NTP time server to the NTP Pool
>>> project and their accuracy, stability and ease of use are just
>>> incredible. The time quality is so good that ntpd even far away
>>> (network wise) usually pick them despite having more nearby other
>>> "good" time sources.
>> The problem is not with the sources. The problem is with the computer
>> crystals and changes in temp. Because ntp is so so slow in converging,
>> changes in temp show up at the usec level. Ie, it is easy using even
>> just a GPS18 to get 10usec accuracy. At the 1us level ntp has troubles.
>> (Since I believe most of the readin ability of ntp is at the 1us level,
>> that is probably the goal)
>>> - ask
> If you require accuracy at the microsecond level, put your equipment in
> a controlled environment! If the temperature in the room where the
> equipment resides is an unvarying 68 degrees F or 20 degrees C you will
> get a better and more stable time. This may not be easy to do but it's
> essential for the very best time. Actually just about any temperature
> that will not damage the hardware will work equally well as long as it
> does not vary minute to minute or hour to hour.
Nope. Most of the heat inside the computer ( where the clock crystal is)
comes from the computer itself. I see this-- duing the day twhen there
is more work being done by the computer, the temp goes up. Of course you
could put the crystal into an oven to make sure its temp remains
constant. Or you could use chrony and get fast convergence and far
better stability (it can keep even a normal computer to microsecond
accuracy from say a GPS source).
More information about the questions