First page Back Continue Last page Overview Graphics
Scaling the master node
On our largest cluster, with 400 CPUs, the master node will not suffice alone to manage all services. We have split the work and functions of the master node to a total of eight service nodes.
The users may not login to the master node, but only to one of three separate login nodes, for three separate user groups.
We use four file servers, not counting the master node, using NFS for file sharing. The current file servers are all Linux machines, resembling the compute nodes in their hardware and software configuration.