TABLE OF CONTENTS Another topology for multiprocessor exploitation is to use multiple servers in a cluster. This was described in Chapter 6 in some detail, though the terminology is somewhat vague until now as we mix systems language with database language. Let s try now to distinguish the concepts and clarify. Chapter 6 describes a database concept known as shared-nothing partitioning in which multiple servers (distinct physically or perhaps only logically) are used to each hold and process a fragment of the database. Shared-nothing partitioning refers to how data is divided and handled, and it is a database term. However, in the systems community (nondatabase world) the use of multiple servers working as a group is called server clustering or machine clustering. In the systems world databases are just one possible way that machine clustering can be used. Machine clusters are often used for scale-out (additional performance) but also to provide failover, which will be discussed shortly in this chapter. Machine clustering does not typically include Grid computing since Grid generally assumes a distributed computing model where the servers are geographically spread out possibly throughout a building or a city, and possibly even across international and continental borders.
