The Model Application Cluster (MAC)
The Model Application Cluster (MAC) runs the WRF numerical weather prediction model to create RTFDDA forecasts and sends the resultant imagery to the range DAS systems. Each MAC system consists of 1 admin node, 2 interactive nodes, 14 compute nodes, and a RAID for storage.
The Admin Node
The admin node is the system administrator's gateway into the MAC. Its primary purposes are to control the interactive and compute nodes,for OS level installations and upgrades that are pushed to the interactive and compute nodes, and for monitoring the system during operations.
The Interactive Nodes
There are two interactive nodes on the MAC, known as int1 and int2. These nodes are similar to the master node on the older MAC systems. The MAC also has an internal RAID for data storage attached to either int1 or int2 and cross-mounted to all other nodes. Processing runs on the interactive node that the RAID is physically connected to. The RAID consists of 11 disks (10 in a RAID 5 configuration with 1 hot spare) and a total of 1.2 TB. The interactive nodes control the RTFDDA model runs. If one interactive node crashes, the other node can take over after a cold swap of the RAID and some additional configuration changes.
The Compute Nodes
As mentioned above, there are 14 compute nodes on each MAC, which are analogous to the slave nodes on the older MAC systems. During each model run, the compute nodes' purpose is to grind away on the complex algorithms of the WRF model. To access the compute nodes, the user must log in to either int1 or int2; a user cannot log in directly to the compute nodes. The compute nodes share information via a high-speed networking that runs between the compute nodes and passes model data back and forth, Myrinet currently.