Parallel computers can be classified by instruction and data streams. Another distinction is between shared and distributed memory systems. We define clusters and the scalability of a problem. Network topologies apply both to
hardware configurations and algorithms to transfer data. Three factors drove the clustering revolution in computing. First is the availability of commodity hardware: choice of many vendors for processors, memory, hard drives, etc... Second, concerning networking, Ethernet is dominating commodity networking technology, supercomputers have specialized networks. The third factor consists of open source software infrastructure: Linux and MPI.