In this chapter we will look at both clusters and massively parallel machines, starting with clusters. Both these architectures are often referred to as being loosely-coupled, a term based on the observation that the processors in one node do not have any close link with the processors in another node processors in different nodes do not share memory (in general, although some clusters do support shared memory), and each node runs its own copy of the operating system. Such unshared-memory systems are also sometimes called NORMA, for No Remote Memory Access.
The concept of a cluster was introduced towards the end of the 70s, by Tandem [1] with its Guardian systems, and was further exploited by Digital with its VAXclusters, introduced in 1983 [KRO86]. Interested readers may found more on clusters in following reference book [PFI95]. The use of the term cluster in a computing context is a simple extension of a fine English term (Webster's definition: "cluster n. [Old English clyster] a number of persons or things grouped together vi., vt. to gather or grow in a cluster").
A cluster is fundamentally some reasonable number of systems around ten is a practical maximum interconnected so that they can share resources. We will refer to such resources as cluster resources. Each system in a cluster is called a node. Each node is a real, complete system, with all the resources it needs processor, memory, I/O devices, and storage sub-system working under the control of its own copy...
TABLE OF CONTENTS 
