Apparently, another corporate router CISCO ASA connected to DMZ was the troublemaker. After physical disconnect and reboot of that device, everything started to work fine again :-)

RRD intentionally aggregates data into larger intervals as the data gets older. The monitoring graphs are intended to provide troubleshooting information, not be a high-resolution, historical archive. For that you can query the device using something like cacti or zabbix or a plethora of others. Setting 8 hours x 1 minute resolution is pretty comprehensive. Anything longer than 8 hours and the resolution will be reduced.