2.3 Distributed Component DBMSs

In this section we describe how to modify a single-site ORDBMS to make it work across a network of computers. With distributedORDBMSs we face many of the same challenges as with distributed RDBMSs: How do we ensure transactional guarantees for distributed operations?; How do we present users with a global schema name space while providing for local autonomy?; and How do we optimize queries that require data movement over the network? In addition, a distributed ORDBMS must address a set of challenges that are the consequence of introducing programmatic, user-defined extensions into the data model. Depending on how it is implemented, extension logic may be mobile (which means it can be copied across the network to wherever it makes the most sense to invoke it) or fixed (which means it can be run only on a particular node of the distributed system). Distributed RDBMS products can assume that every node in the system has identical functionality or, in the case of a federation of different RDBMS products, very similar (i.e., mapable) functionality. But with a distributed component DBMS, each node might contain disjoint sets of component extensions.

Fortunately, many of the mechanisms developed for distributed data management in RDBMSs can also be applied in distributed ORDBMSs. For example, the theoretical models and engineering techniques used to implement distributed transactions can be reused. Therefore, we focus our attention here on topics related to query processing rather than lower-level data management. Also, we assume that within our distributed...