¡Esta es una revisión vieja del documento!
Tabla de Contenidos
Sincronice su red con NTP
NTP ( Network Time Protocol ) permite la sincronización de reloj entre sistemas informáticos. El siguiente HOWTO describe:
- configurando un servidor NTP en Slackware Linux;
- Sincronizar PCs cliente con su servidor NTP local.
Introducción
When several users manipulate shared data on different client PCs on a network, it's important that these machines are all synchronized. This is especially true if you share files over NFS, or if you use NIS for centralized authentication. You'll get all sorts of weird errors if your clocks are out of sync. Unfortunately, the clients' onboard clocks aren't sufficiently precise.
That's where NTP (Network Time Protocol) comes in handy. It allows networked machines to adjust their clocks so as to be perfectly synchronized. A series of public time servers on the Internet allow the reception of the exact time. From this point, we can use NTP in several ways.
- The
ntpdate
command makes an initial correction of the BIOS clock. - This one-time-adjustment isn't sufficient for a server that is supposed to be up 24/7, since its clock will drift away gradually from the exact time. In that case, we have to configure the
ntpd
daemon (shipping with thentp
package). This daemon contacts public time servers at regular intervals and proceeds with incremental corrections of the local clock. - The
ntpd
daemon can in its turn be configured as a time server for the local client machines.
It's considered good practice to use ntpdate
for the initial adjustment and ntpd
for regular time synchronization.
Firewall considerations
The NTP services uses UDP port 123. Open this port if you want to allow remote machines to connect to your NTP server.
Synchronize a LAN server or a public root server with an NTP server on the Internet
Create an empty log file:
# touch /var/log/ntp.log
Visit http://www.pool.ntp.org and choose a list of servers according to your country.
Configure the NTP service by editing /etc/ntp.conf
. You might backup the existing ntp.conf
file and start from scratch.
In the example below, the list of four servers is chosen for my company's location (France):
# /etc/ntp.conf driftfile /etc/ntp/drift logfile /var/log/ntp.log server 0.fr.pool.ntp.org server 1.fr.pool.ntp.org server 2.fr.pool.ntp.org server 3.fr.pool.ntp.org server 127.127.1.0 fudge 127.127.1.0 stratum 10 restrict default nomodify nopeer notrap restrict 127.0.0.1 mask 255.0.0.0
Here's a little explanation for some options:
- The
fudge 127.127.1.10 stratum 10
directive is a “dummy” server acting as fallback IP in case the external time source becomes momentarily unreachable. When this happens, NTP will continue to work and base itself on this “internal” server. - NTP has its own arsenal of rules to limit access to the service, which can be used independently from a firewall. The
restrict
directives in the above configuration prevent distant computers from changing the servers' configuration (firstrestrict
statement), and the machine is configured to trust itself (secondrestrict
statement). - A
restrict
statement without any argument but followed by the hostname boils down to anallow all
.
Manage the NTP service
Before starting the service, proceed to an initial adjustment of your system clock:
# ntpdate pool.ntp.org
ntpdate
command is normally considered obsolete, but it still comes in handy when performing important time adjustments. The “orthodox” way would be to use the ntpd -g
command - the official replacement for ntpdate
- but its use will fail if your system clock is off for more than half an hour.
Activate the NTP service:
# chmod +x /etc/rc.d/rc.ntpd
Manage the NTP service:
# /etc/rc.d/rc.ntpd start|stop|restart|status
Now display the list of servers your machine is actually connected to:
# ntpq -p remote refid st t when poll reach delay offset jitter ============================================================================== *panopea.unstabl 213.251.128.249 2 u 30 64 377 56.136 -249.48 80.680 +88-190-17-126.r 145.238.203.14 2 u 29 64 377 77.571 -205.94 94.278 +62.210.255.117 192.93.2.20 2 u 29 64 377 77.097 -249.57 85.641 -ntp.univ-poitie 145.238.203.10 3 u 29 64 377 57.747 -191.58 107.002 LOCAL(0) .LOCL. 10 l 164 64 374 0.000 0.000 0.001
The little *
asterisk preceding one of the above lines means your machine is effectively synchronized with the respective NTP server.
Synchronize your client PC(s) with your local NTP server
In a LAN, it is considered good practice to synchronize only one machine - the server - with a public NTP server, and the client PCs with the local server. This saves bandwidth and takes some load off the public NTP servers.
As above, proceed to an initial adjustment of the system clock:
# ntpdate pool.ntp.org
Create an empty logfile:
# touch /var/log/ntp.log
Now configure NTP to synchronize with the LAN server. Replace the example's IP (192.168.2.1
) with your real server's IP:
# /etc/ntp.conf driftfile /etc/ntp/drift logfile /var/log/ntp.log server 192.168.2.1 server 127.127.1.0 fudge 127.127.1.0 stratum 10 restrict default ignore restrict 127.0.0.1 mask 255.0.0.0 restrict 192.168.2.1 mask 255.255.255.255
- The three
restrict
statements mean we're blocking all NTP traffic except for the client itself and the server.
Activate and start the NTP service:
# chmod +x /etc/rc.d/rc.ntpd # /etc/rc.d/rc.ntpd start
As above, use the ntpq -p
command to check if the synchronization went well:
# ntpq -p remote refid st t when poll reach delay offset jitter ============================================================================== *192.168.2.1 81.19.16.225 3 u 916 1024 377 0.367 7.897 2.552 LOCAL(0) .LOCL. 10 l 10h 64 0 0.000 0.000 0.000
Monitor the performance of ntpd
You will notice that the logfile /var/log/ntp.log
does not contain any information about the actual accuracy of your system clock. If it's important to you, you can log the statistics of time corrections applied by NTP daemon to the system clock. To do this, add the following lines to /etc/ntp.conf
:
statsdir /var/log/ntp/ statistics loopstats filegen loopstats file loops type day link enable
You have to create the statsdir manually. Once the configuration changes are in effect, ntpd
will create files named loops.YYYYMMDD
in that directory. Below is an example line from one of these files:
56690 3950.569 0.001199636 2.297 0.001830770 0.571576 10
The first and second number are the UTC time (expressed as Modified Julian Date and seconds elapsed since midnight). The third and fourth number are the offsets of time (in seconds) and of frequency (in parts per million). The fifth and sixth number are their respective uncertainties.
To monitor the performance of ntpd
, you can examine the plot of clock offset or frequency offset vs. time:
$ awk '{printf("%f %f %f\n", $1+$2/86400, $3, $5)}' /var/log/ntp/loops.* > time $ awk '{printf("%f %f %f\n", $1+$2/86400, $4, $6)}' /var/log/ntp/loops.* > freq $ gnuplot gnuplot> set xzeroaxis gnuplot> plot 'time' with yerror gnuplot> plot 'freq' with yerror
Given enough data, visual examination of the plots will allow you to see peculiarities in ntpd
performance, should they arise. For example, in the case illustrated by the figure below, the rapid decrease of the frequency offset was caused by replacing the power supply unit of the machine.
Sources
- Originally written by Niki Kovacs
- Performance monitoring section contributed by Dominik Drobek