An architecture for object replication in distributed systems /

Replication of objects (services) is one way of providing fault tolerance in distributed systems where multiple instances of the objects are executed on machines that fail independently. The copies of the objects are kept in a consistent state with the help of a replica consistency protocol. For t...

Full description

Bibliographic Details
Main Author: Beedubail, Ganesha Rao
Format: Thesis Book
Language:English
Published: [Place of publication not identified] : [publisher not identified] ; 1997.
Subjects:
Online Access:http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=739891211&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD
Description
Summary:Replication of objects (services) is one way of providing fault tolerance in distributed systems where multiple instances of the objects are executed on machines that fail independently. The copies of the objects are kept in a consistent state with the help of a replica consistency protocol. For the user (client) the replicas appear as a single object. This notion is called replica transparency. The objectives of this research are providing an architecture for developing replicated objects and developing efficient algorithms for replica consistency and management in object oriented distributed systems. However, the resulting architecture is general and the concepts can be adapted to non-object oriented distributed systems. The architecture uses only standard off-the-shelf hardware and software and greatly reduces the complexity of the development of replicated objects in distributed systems. The architecture specifies the main requirements for developing replicated objects in distributed systems, namely: (a) the replicated object representation, (b) the replicated object development and management mechanism, and (c) the replica consistency protocol. For achieving these requirements the architecture specifies a hierarchy of objects and the inheritance and functional relationship among these objects. All the management functions of replicated objects are encapsulated in a manager object. An application object inherits these functionalities from the manager object. The client accesses the replicated object through a proxy which makes the replication transparent to the client. This architecture can also be effectively adapted to wrap legacy objects for making them fault tolerant. The architecture can support any replica consistency protocols. However, this dissertation presents an efficient replica consistency algorithm that takes advantage of the features of the replication architecture. It is less expensive compared to the existing replica consistency protocols. This replica consistency protocol is called the hot replication protocol. This protocol is a modification of the primary-backup (master-slave) protocol in which the slave replicas are always synchronized with the master. Hence they can assume the role of master with minimal delay when the current master fails. The failure detection and group membership protocols are seamlessly integrated in the replica consistency protocol. An analysis of the replica consistency protocol is also carried out and it is compared with some of the existing replica consistency protocols. A prototype implementation of the architecture has been carried out in Spring OS environment. The performance results are presented and discussed.
Item Description:Vita.
"Major Subject: Computer Science".
Physical Description:xi, 121 leaves : illustrations ; 28 cm.
Issued also on microfiche from University Microfilms Inc.
Bibliography:Includes bibliographical references: pages 113-120.