[ntp:questions] Re: Do we need / should be have a refclock?

Richard B. Gilbert rgilbert88 at comcast.net
Sat Jul 8 16:56:57 UTC 2006

Robert Hulme wrote:

> Hi,
> I'm fairly new to NTP (although I have read a lot of the documentation
> on the site and in the wiki) and I'm trying to get some expert advice to
> help with an internal IT discussion occurring in the IT team at my
> company.
> We have about 20-30 linux servers collocated (on two sites - 95% of them
> on one site) and about 30 windows desktops at a different site. The new
> Linux sysadmin has noticed that the NTP client configuration on the
> servers is fairly random (they seem to use various random UK ISP's NTP
> servers (only one NTP server configured for each box)). Clearly that
> needs sorting out - but in the course of the discussion there was a
> debate about whether we ought to get a refclock.
> Could you give some advice on how we should be thinking about this, what
> we should consider, and what the pros and cons of using a refclock are.
> For reference we basically do web hosting for our clients on these
> servers - some run databases and so on but there is nothing particularly
> unusual that the servers do.
> If you think a refclock is a good idea or an important thing to have the
> one that has been suggested is this one
> http://www.atomic-clock.galleon.eu.com/Atomic-Clock-Time-Server/Atomic-C
> lock-Time-Server.htm
> If on the other hand there is a simple place I can RTFM then please
> point me to it :-)
> Robert Hulme
> Technical Consultant
> Anthony Hodges Consulting Limited
> Direct line: (01924) 203904
> Switchboard: (01924) 203900
> Email: Robert.Hulme at ahc.uk.com
> _______________________________________________
> questions mailing list
> questions at lists.ntp.isc.org
> https://lists.ntp.isc.org/mailman/listinfo/questions

You really need to define your requirements.  When you have specified 
the accuracy you need and the reliability you need, we can probably give 
better advice.  If traceability of your timestamps to a national 
standards lab is a requirement be sure to mention that as well

A refclock, particularly a GPS timing receiver, provides an ultra stable 
and ultra accurate reference.  The stability means that clients can 
synchronize with the server synchronized to that refclock and stay 
within, say, 50 microseconds of it.  A great deal depends on the 
hardware and O/S being used.  A refclock can, but need not, be 
expensive; mine cost me $200 US.  The cheapest I know of is about $85 US.

The fast, easy, and cheap solution, is to designate one of those Linux 
boxes as your house timeserver and configure it to use four or five 
internet servers.  Then tell all the other boxes to synchronize with it.
You can expect everybody to stay within ten to twenty milliseconds of  UTC.

If you need better reliability than a single in house server can 
provide, you can use more.  That is a good reason to define your 
requirements up front.

More information about the questions mailing list