Partitioned operation and distributed data base management system catalogs
Greg Thiel · 1983
Currently the cost of computer hardware is rapidly decreasing and the hardware technology of local networks is rapidly expanding and decreasing in cost. These facts, coupled with the observation that at a major portion of all computing is data base related activity, indicate that distributed data base systems will become an essential part of the computing done in the future. One important set of system related questions are those that relate to separating the lowest levels of a distributed data base system from the actual data base operations and making these lower levels part of a distributed operating system. As one example of this problem the design and implementation of a DDBMS's name space and naming mechanism is done. Careful consideration is given to where functionality should be provided to minimize redundant work. A second important set of questions concerns replication and partitioned operation. Replication of data in a data base presents the opportunity to provide users with increased reliability and availability. The central question is how to provide the increased reliability and availability in the face of failures. Previous work done at UCLA is expanded and enhanced to overcome multiple implementation issues. As a naming mechanism case study the INGRES Data Base System and the LOCUS Operating System (developed at UCLA) will be examined from the context of architectural issues. Due to the network transparency of LOCUS the standard centralized INGRES system can already run on LOCUS with distributed and replicated data. All of these issues will grow in importance as the prospect of a computer in every home ceases to be a dream.