Jamie told me to use chroot to log in into the chroot jail of debian os that are exported for the FEs and install ntp there. I took following steps at the end of which, all FEs have NTP synchronized now.

• I logged into fb1 through nodus.
• chroot /diskless/root.jessie /bin/bash took me to the bash terminal for debian os that is exported to all FEs.
• Here, I ran sudo apt-get install ntp which ran without any errors.
• I then edited the file in /etc/ntp.conf , i removed the default servers and added following lines for servers (fb1 and nodus ip addresses):
  server 192.113.168.201
  server 192.113.168.201
• I logged into each FE machine and ran following commands:
  sudo systemctl stop systemd-timesyncd.service; sudo systemctl status systemd-timesyncd.service;
  timedatectl; sleep 2;sudo systemctl daemon-reload;  sudo systemctl start ntp; sleep 2; sudo systemctl status ntp; timedatectl
  sudo hwclock -s
  • The first line ensures that systemd-timesyncd.service is not running anymore. I did not uninstall timesyncd and left its configuration file as it is.
  • The second line first shows the times of local and RTC clocks. Then reloads the daemon services to get ntp registered. Then starts ntp.service and shows it's status. Finally, the timedatectl command shows the synchronized clocks and that NTP synchronization has occured.
  • The last line sets the local clock same as RTC clock. Even though this wasn't required as I saw that the clocks were already same to seconds, I just wanted a point where all the local clocks are synchronized to the ntp server.
• Hopefully, this would resolve our issue of restarting the models anytime some glitch happens or when we need ot update something in one of them.

Edit Tue Aug 24 10:19:11 2021:

I also disabled timesyncd on all FEs using sudo systemctl disable systemd-timesyncd.service

I've added this wiki page for summarizing the NTP synchronization knowledge.