Above, we discussed carefully designed systems with special high-speed networks linking nodes usually derived from commercial PC or workstation technology. However, many have explored using workstations linked with conventional networks, such as Ethernet, fddi, or ATM as an informal parallel system. Such COWs or NOWs (clusters or networks of workstations) are clearly MIMD-NUMA distributed memory parallel machines. They should be able to run any software or parallel algorithm designed for machines such as the IBM SP2 or Intel Paragon. However, the latter machines have higher bandwidth and lower latency communication. Thus, they can support many more applications with high efficiency. However COWs are attractive because experience has shown that they do perform many parallel tasks effectively. They offer many political and fiscal advantages. COWs can be formed from existing ``idle'' workstations and PCs overnight and on weekends. In principle, no cost is involved other than that already incurred for desktop computers to support the organization.
These ideas can be generalized to metacomputing where any arbitrary set of heterogeneous computers are linked together to address a single problem.