Clusters and Grids

Abstract

AbstractOver the past 10–15 years,cluster computinghas emerged as a viable alternative to MPPs by offering a significantly superior price/performance ratio. Normally, a cluster consists of hundreds or thousands of personal computers (PCs) or workstations interconnected through commodity network cards. One can recognize the difference between a cluster and an MPP similar to that between a mainframe computer and a Personal Computer (PC). Both possess a Central Processing Unit (CPU), random access memory, disks, one operating system, and so on. On the mainframe, everything works a lot faster than on the PC, but it lags behind in terms of quality and requires additional effort and cost for maintenance and administration. The same holds true between MPPs and clusters. Originally, MPPs excelled mostly because of the proprietary high‐performance interconnections between standard processors, such as Intel Pentium, Sun UltraSPARC, or PowerPC. Following the analogy with mainframes being replaced by PCs, cluster architectures are nowadays a viable replacement for MPPs, the driving force that enabled this transition being the rapid improvement in the availability of commodity high‐performance components for PCs/workstations and networks. These technologies are making a network/cluster of computers an appealing vehicle for cost‐effective parallel processing, and it is consequently leading to low‐cost commodity supercomputing.