Time synchronization service and why choosing ntpd on a Pogoplug E02

What is NTP ?

On any computer system, having a precise time set is crucial for running it. This is specially so for the Pogoplug E02 who doesn’t have a dedicated hardware RTC (real-time clock)¹ to keep track of the current time. This means that after a reboot, the time is reset until it’s set manually. But because there is no hardware to maintain its correctness, we must use a time synchronization service to keep it in check. A protocol being used for that is called NTP (Network Time Protocol)², who uses external services to set proper time in intervals.

NTP daemons

For Linux systems there are various implementations of software to use the NTP protocol, one of them is actually part of the systemd³ suite that is packaged with an Arch Linux installation. It’s called systemd-timesyncd⁴.

It’s configured in /etc/systemd/timesyncd.conf and you can run it with the systemctl command as following.

$ systemctl start systemd-timesyncd

But as we’re using the Arch Linux installation on an ARM device which uses an USB flash storage to keep the system, it’s beneficial to keep as less as possible write cycles to the system log as possible, so the storage doesn’t degrade that fast. The problem with the systemd-timesyncd daemon is that it writes too much log data often with no way to stop.

We can see that with using the system log viewer command journalctl and filtering the output with grep. -b switch displays whole log since boot.

$ sudo journalctl -b | grep timesyncd
Aug 28 04:24:43 pogo systemd-timesyncd[105]: Using NTP server 89.212.75.6:123 (0.arch.pool.ntp.org).
Aug 28 04:26:03 pogo systemd-timesyncd[105]: interval/delta/delay/jitter/drift 2048s/+0.001s/0.066s/0.002s/+27ppm
Aug 28 05:00:11 pogo systemd-timesyncd[105]: interval/delta/delay/jitter/drift 2048s/-0.002s/0.061s/0.003s/+26ppm
Aug 28 05:34:19 pogo systemd-timesyncd[105]: interval/delta/delay/jitter/drift 2048s/+0.002s/0.066s/0.002s/+27ppm
Aug 28 06:08:27 pogo systemd-timesyncd[105]: interval/delta/delay/jitter/drift 2048s/-0.002s/0.063s/0.003s/+26ppm

It’s obvious that using systemd-timesyncd flash storage will have to write heavily throughout the day and that doesn’t sound good if we intend to keep the same USB key for a year or more. But there’s an alternative service for NTP synchronization.

ntpd

Ntpd⁵ is an older NTP daemon, who does the same job. Installing it on an Arch Linux installation is a breeze using the built in pacman⁶ package manager as we worked with it before in the initial installation post.

At first we stop and disable the systemd-timesyncd daemon to keep it running at boot.

$ sudo systemctl stop systemd-timesyncd
$ sudo systemctl disable systemd-timesyncd

Then we install the ntp⁷ package with pacman, after updating the package source database first using the -Sy parameter.

$ sudo pacman -Sy
$ sudo pacman -S ntp

The configuration for it is saved in /etc/ntp.conf, with an option to change the desired NTP time servers and restricting where the queries may originate.

  1 # Please consider joining the pool:
  2 #
  3 #     http://www.pool.ntp.org/join.html
  4 #
  5 # For additional information see:
  6 # - https://wiki.archlinux.org/index.php/Network_Time_Protocol_daemon
  7 # - http://support.ntp.org/bin/view/Support/GettingStarted
  8 # - the ntp.conf man page
  9
 10 # Associate to Arch's NTP pool
 11 server 0.arch.pool.ntp.org
 12 server 1.arch.pool.ntp.org
 13 server 2.arch.pool.ntp.org
 14 server 3.arch.pool.ntp.org
 15
 16 # By default, the server allows:
 17 # - all queries from the local host
 18 # - only time queries from remote hosts, protected by rate limiting and kod
 19 restrict default kod limited nomodify nopeer noquery notrap
 20 restrict 127.0.0.1
 21 restrict ::1
 22
 23 # Location of drift file
 24 driftfile /var/lib/ntp/ntp.drift

Default configuration is pretty good, so we can keep it as default. Now we just have to start the ntpd daemon.

$ sudo systemctl start ntpd

When checking the log file, we see that ntpd doesn’t log the synchronization queries at all, only starting and stopping the service.

$ sudo journalctl -b | grep ntpd
Aug 28 17:07:39 pogo ntpd[15780]: ntpd 4.2.8p1@1.3265-o Fri Feb  6 18:06:44 UTC 2015 (1): Starting
Aug 28 17:07:39 pogo ntpd[15780]: Command line: /usr/bin/ntpd -g -u ntp:ntp
Aug 28 17:07:39 pogo ntpd[15781]: proto: precision = 2.385 usec (-19)
Aug 28 17:07:39 pogo ntpd[15781]: Listen and drop on 0 v6wildcard [::]:123
Aug 28 17:07:39 pogo ntpd[15781]: Listen and drop on 1 v4wildcard 0.0.0.0:123
Aug 28 17:07:39 pogo ntpd[15781]: Listen normally on 2 lo 127.0.0.1:123
Aug 28 17:07:39 pogo ntpd[15781]: Listen normally on 3 eth0 192.168.1.2:123
Aug 28 17:07:39 pogo ntpd[15781]: Listen normally on 4 lo [::1]:123
Aug 28 17:07:39 pogo ntpd[15781]: Listen normally on 5 eth0 [fe80::225:31ff:fe04:9b90%2]:123
Aug 28 17:07:39 pogo ntpd[15781]: Listening on routing socket on fd #22 for interface updates

At the end, we set it to be run at boot and to update the time at network connection.

$ sudo systemctl enable ntpd
$ echo "ExecUpPost='/usr/bin/ntpdate -u pool.ntp.org'">>/etc/netctl/eth0

Conclusion

As demonstrated above, running a Linux server with various services on a smaller ARM device with limited resources and capabilities requires a specialized mindset towards choosing those services, so we have find some alternatives that are better in the long run.