[mosca]

Reading about it, it looks like it could combine multiple sgi origin 300 as a single machine with with NUMA Link cables. That way it would probably behave as a single bigger machine and you could just run a multithreaded application or multiple compute cores on what looked like a much bigger machine.

It would likely be connected to a big storage array for input/output data and the network would be used for control. IPC accross nodes would happen over the much faster NUMA Link.

You would likely use something like MPI to write the distributed compute tasks but there were different frameworks for scheduling and controlling HPC jobs across multiple nodes over the network.