[ntp:questions] How to keep Linux server in Chicago and Mumbai in sync to within 5 microseconds

Chris Albertson albertson.chris at gmail.com
Wed Jan 12 05:06:41 UTC 2011


>> Also the OP is asking for +- 5 microseconds.  That is moderately hard.
>
> CDMA is outside of that ...

Good catch on that.   But I think it is not a CDMA vs. GPS thing.
turns out if you buy an EndRun CDMA box your Linux server then is an
NTP stratum 2 client with the EndRun acting as a Stratum 1.    Inother
words in EndRun's plan the time is sent over the network to the Linux
box.    But with GPS your Linux box is the itself the root level
server and there is no network delay.

The requirement the the two Linux server be no more than 5 uS apart
means each must be within 2.5 uS is "true".  I think (?) this means
that both Linux serversw ill need a reference clock them selves

I'm faily certain there are CDMA reference clocks that connect
directly to Linux servers.  What you need inside ssuch a clock is a
good local oscillator that is CDMA disciplined and then divided down
to one pulse per second.  I'm pretty sure I've seen these

I'm setting up one of there right now, recompiling the latest NTP (but
not having yet having desired result, can't seem to enable the Oncore
ref clock, forced to use NEMA) I'm also experimenting with two kinds
of antenna and a few locations and different cable runs.
Conclusions so far:  (1) GPS can be "way low cost"  using $15 Oncore
UT+ GPS unit and $25 Lucent antenna.   (2) cable type matters if the
run is 25+ feet (3) View of sky is reallt required if location
uncertianty is to be removed.  It is well worth installing a mast to
get above local obstructions


*** Next question for the OP ***
So you have read all this and decided to hire a consultant to buy and
set up the required equipment.  Before you write him that last big fat
check how will you test his setup to make sure you are even within a
factor of 1000 of your required performance.    With zero hardware,
using just pool servers on the 'net I could get your two servers
within a few milliseconds  of each other.  That is 1000 times worse
then you paid him for but how would you know?  Yes there are ways but
how would YOU know?  If you can't answer this then maybe don't pay
vendors big sums of money.  Do it yourself, start with using pool
servers, then set up your own reference clocks, yes with an "s'"
plural of "clock".  Without multiple sources you will never know if
one is wrong.  I wrote above that this is a moderately hard problem.
Geting there is a process that will take time, not just plug in a box
and click a mouse.It need not be an expensive process

I'm still wanting to know the reason for this.  If we knew the big
picture then maybe some one could suggest a way to solve the problem
in a simpler way.  For example in our business we send IRIG time code
along with the data and so don't need to keep computers time in sync,
So there may be some way to side step the problem

-- 
=====
Chris Albertson
Redondo Beach, California



More information about the questions mailing list