« previous 1 2 3 4 next »

Server selection

Initially, the administrator can enter some of public servers such as ntp.cis.strath.ac.uk at the University of Strathclyde, Glasgow or Microsoft's time.windows.com. It is important when doing this to adhere strictly to access rules. That means only using servers when your computer is within their service area and registering by email if it is required by the server operator.

Once you have started the server process and had a nice long coffee break, it is time to take a look at the NTP server evaluations. To do this, use the ntpq -p command, which gives output such as excerpted below (the listing has been shortened for formatting clarity). The servers that have a zero in the "reach" column could not be reached by ntpd. As long as there are no fundamental network problems, these servers can be deleted from the configuration.

[root@timeserver]# ntpq -p
     remote	      refid    st t when poll reach   delay   offset 
=====================================================================
 LOCAL(0)	 LOCAL(0)      10 l    3   64	 3    0.000    0.000  
 ntp0.freenet.de 0.0.0.0       16 u    -   64	 0    0.000    0.000  
*ntp1.ptb.de	 .PTB.		1 u   31   64	 1   44.181   -0.802  
-chronos.cru.fr  .GPS.		1 u   17   64	 1   52.320   -4.261  
+ntp-p1.obspm.fr .1PPS. 	1 u   15   64	 1   49.748   -1.492  
 time.heise.de	 ntp1.ptb.de	2 u   50   64	 7    0.502   -1.334   

The "refid" indicates what kind of time source the server itself uses. Stratum 1 servers display .GPS. for a GPS receiver, .DCF. for a DCF77 radio clock and .PPS. for a clock that generates one pulse per second. Higher strata servers display their reference server as the refid.

The first character of the first column, in front of the server name, is the ntpq synchronisation status code: a star indicates the current reference server, a plus sign denotes servers that are used in the mean time calculation, a # indicates a server that would also be acceptable, but is further back in the evaluation order. A computer that does not deliver any dependable values for a long period of time is marked with an x (falseticker). Servers that ntpq marks with a minus sign are those that deviated too much from the reference time the last time they were polled. If this occurs frequently in a fairly distant server it should be deleted from the list because it could also be a falseticker that ntpd cannot detect on its own.

Querying with ntpq should be repeated once ntpd has had several hours to repeat its evaluations. If at this point the list has not shrunk down to between six to ten servers, you should delete some of the servers that are further away in terms of internet topology, because each polling generates network traffic. Repeatedly polling computers that never make it into the top ten that are marked with a plus sign wastes resources both on the network and on your own computer. At least three servers should remain, however, so that in case one of them fails, the local server does not have to rely on the inaccurate internal clock, and so that ntpd has an adequate basis for evaluation.

« previous 1 2 3 4 next »