The network problem affected many hosts on the site access network switches and was caused by the main radius server not working properly because of the filling of its system disk.
By configuration, for the majority of the hosts the network connections already up should not have been dropped, irrespectively of the radius server status. Why this didn't happen is under investigation.