Search

As the computer industry has grown in experience and application, there is an area which has received much attention; that is, the area of managing, organizing and storing data. Problems of reliability, security and timeliness with data have drawn almost every computer user's attention. Since the late 60's, there has been a growing interest in providing software which separates (at a cost) the manipulation aspect from usual organizational and accounting functions found in a normal program. Such 'housekeeping' chores can require significant efforts on the part of programmers. Further, once such chores are initially done they will have to be redone time and again, since requirements and usage will change. (Author)

One of the major obstacles to the widespread development and utilization of distributed data base management systems is the lack of an efficient recovery technique. A methodology is presented here for recovery of distributed data bases. The central operation of the recovery technique is rollback of a data base application task on the processor which controls access to the data. The rollback procedure restores the data base to its original state prior to the execution of the application task and determines the set of applications tasks which may have been effected by that task. Tasks that have not operated upon data altered by tasks being rolled back are not affected by the procedure. The rollback procedure attempts to minimize the time and space requirements for recovery. (Author), See also Volume 18, AD-A103 100.

This paper describes a data base management system under development at Kansas State University, intended for use in a network composed primarily of minicomputers. The report presents a description of the computers forming the network and their intercomputer communication system. The data base management system is a network type as specified by CODASYL. An extension of a CODASYL-type DBMS to multicomputer configurations is presented and several DBMS network topologies are discussed. We then conclude with a discussion of a completely distributed data base network. (Author), See also Volume 17, AD-A103 101.

This document is a proposal for a distributed data base management system (DDBMS). It represents the first phase of the DDBMS design portion of Grant 108. It is very important to note that this document is a proposal and also that the next phase of the design is the development of the functional specifications of the DDBMS. Therefore, it is essential that all interested parties respond with any corrections, additions, deletions, suggestions, etc. by July 1, 1976. As it can easily be observed from this report, the implementation of the complete DDBMS will be an enormous task. Estimates range from 7 to 20 up to 50 person years of effort. A natural course is to design a full scale system and proceed with the implementation in an incremental manner. The implementation of a minimal prototype should be achieved as soon as possible for purposes of feasibility studies, testing, and morale., See also Volume 19, AD-A103 102.

The problem of deadlock in distributed data base management is analyzed in terms of performance effects of potential deadlock handling schemes. The performance tradeoffs of deadlock detection and deadlock prevention for distributed data base management systems are compared. Since the run-time overhead in deadlock prevention is projected to be less than for deadlock detection, an algorithm for preventing deadlocks in distributed data base systems is developed. The critical information for the deadlock prevention algorithm is maintained in a shared record list. The shared record list contains all shared access records for a set of tasks. Shared records lists are maintained dynamically by the run-time system. A proof that the algorithm prevents deadlocks in a distributed data base management system is provided along with a comprehensive example. (Author), See also Volume 9, AD-A103 074.

In this paper we present an overview of data base management systems (DBMS), the motivation for distributed data base systems (DDBMS), a set of possible network topologies served by the distribution, the mechanisms necessary to integrate (and communicate between) the DDBMS system elements when distributed across a nonhomogeneous network of minicomputers, and some implementation details on a prototype system. The current prototype distributes the DBMS and application program function across an IBM 370/158 and a (minicomputer) NOVA 2/10. In the near future, a third machine, the Interdata 85 minicomputer, will be added to the network. The DBMS used is a network system as specified by CODASYL. The emphasis in this paper will be on the problems posed by the heterogeneous machines and the intertask (processor) communication system which is utilized in the distribution of data, programs, and control. (Author), See also Volume 18, AD-A103 100.

The results of a simulation study intended to determine the circumstances under which it is beneficial to operate a data base management system with a multi-processor backend are presented. The basic concept of backend data base management systems and multi-processor backend systems are provided as background material. The general structure of the simulation model which has been implemented in GPSS is outlined. The results of the study indicate that the amount of CPU activity required by the data base management system is a determining factor with respect to the need for a multi-processor backend. (Author), See also Volume 16, AD-A103 099. Pub. in Proceedings Conference on Information Sciences and Systems, Baltimore, MD Apr 77.

With the advent of Data Base Management Systems (DBMS) and associated facilities (data dictionaries, query languages, report writers, etc.), the task of data organization, management, and storage has been given to a select group of specialists. These specialists (the Data Base Administrators (DBA) provide the necessary control, logging, and access information and software to the program. Such activity relieves the programmers of this overhead function allowing them to concentrate on the necessary manipulations. This paper focuses on some alternatives with respect to a DBMS in terms of a centralized versus decentralized environment. The first section deals with the concepts and tradeoffs involved in considering the two environments. The second section then deals with problems which are encountered in a distributed data base management system. These problems include deadlock, rollback and recovery, data conversion, redundancy, and communication and operating system requirements for effective distribution., See also Volume 15, AD-A103 098.

The combination of general purpose data base management systems and computer networks to produce distributed data base management systems should be the next significant step in commercial systems development. A completely generalized distributed data base management system would reside on a heterogeneous computer network with different data base systems available at various processors. Communication and data transfer would be possible between any nodes in the network. This report described the principal problem areas in distributed data base management system development. Distributed data base systems share many design problems with both single machine data base systems and computing networks, as well as introducing several new dilemmas. Recent research in these problem areas is presented to provide a picture of the state of the art of distributed data base development. In addition, the current status of the data base industry with respect to distributed processing is evaluated by reporting the current projects and future plans of selected (anonymous) data base vendors., See also Volume 12, Ad-A103 097.

A memory management scheme which incorporates an additional level of memory into the traditional primary-secondary storage hierarchy is proposed for utilization in distributed data base management systems. In this scheme, the memory of the back-end processor is used as an additional memory buffer. An optimal three-level memory management algorithm is presented along with an analysis of its cost in terms of page replacement. The expected performance improvement over the optimal algorithm for a two-level memory system is determined. The performance benefits of the three-level memory management are applicable to most distributed processing systems. (Author), See also Volume II AD-A103 096.

A software organization is presented to provide for data definition and manipulation in a distributed data base management system. With the mechanism for distributing the data base proposed here, the physical location of the data is transparent to the user program. A Device Media Control Language is specified for the assignment of control of and access to a data base area to a set of processors. Procedures for reassignment of the control and access functions as well as the transfer of data between processors are provided. The basic hardware and software requirements for a computer network capable of supporting a distributed data base management system are discussed along with a specification of the software required for a processor in a distributed data base network. (Author), See also Volume 8, AD-A103 067.