The NOvA experiment's networks are broken into multiple segments, each servicing a different aspect of the experiments needs. The general design of all of these networks are that they are split generally into Public and Private segments depending on the hardware and functionality that needs to be available on each. They are again divided between the general classifications of the high speed Data Acquisition (DAQ) and lower speed Detector Controls System (DCS) network in order to segregate data and controls traffic.
The DAQ networks are designed to carry the data and controls traffic between the acquisition hardware and the computing farm that is used to buffer and process the data. The networks are sized to accommodate the required connections for roughly 200 data concentrator modules (DCMs) and a corresponding set of 200 farm/buffer computers. The network also supports the timing system and other acquisition specific hardware.
Public DAQ Network¶
The public DAQNet is designed to carry the majority of the low bandwidth command/controls traffic between different components of the NOvA computing clusters. This traffic includes messages sent by and to the run control and other DAQ servers as well as traffic related to the distribution of network file systems and other infrastructure (i.e. dhcp and bootp requests, nfs mounts, snmp monitoring, kerberos authentication, ganglia monitoring and http traffic).
Note: The Public DAQNet must be assigned to a routable IP range to allow for strong authentication with the Kerberos system. Kerberos will not operate correctly with private IP addresses which cannot be resolved through a DNS server.
Currently the Public DAQ network is assigned to the IP address range: 18.104.22.168/24
More information on the public daq network can be found here:
Private DAQ Network¶
The private DAQNet is designed to carry the majority of the actual detector data flow between the DCMs and the buffer nodes. The network also carries a limited amount of messaging traffic to the DCMs. This traffic is highly asymmetric and flows out of the DCMs and into the buffer nodes.