leff@smu.UUCP (Laurence Leff) (05/23/89)
The following technical reports on replicated data managements are available from UC Santa Cruz. Please address correspondence to: Technical Report Librarian Baskin Center for Computer Engineering & Information Sciences Applied Sciences Building University of California Santa Cruz, CA 95064 TR #88-35 Cost $4 The Effect of Failure and Repair Distributions on Consistency Protocols John L. Carroll San Diego State University Darrell D. E. Long University of California, Santa Cruz Abstract The accessibility of vital information can be enhanced by replicating the data on several sites, and employing a con- sistency control protocol to manage the copies. Various protocols have been proposed to ensure that only current copies of the data can be accessed. The effect these protocols have on the accessibility of the replicated data is investigated by simulating the operation of the net- work and measuring the performance. Several strategies for replica maintenance are considered, and the benefits of each are analyzed. The details of the simulations are discussed. Measurements of the reliability and the availability of the replicated data are compared and contrasted. The sensitivity of the Available Copy and Dynamic- linear Voting protocols to common patterns of site failures and repairs is studied in detail. Exponential, Erlang, uni- form, and hyperexponential distributions are considered, and the effect the second moments have on the results is analyzed. The relative performance of competing protocols is shown to be only marginally affected by non-exponential distributions, validating the robustness of the exponential approximations. TR #88-34 Cost $4 Reliability of Replicated Data Objects Darrell D. E. Long University of California, Santa Cruz Jehan-Francois Paris University of Houston John L. Carroll San Diego State University Abstract Improved fault tolerance of many applications can be achieved by replicating data at several sites. This data redundancy requires a protocol to maintain the consistency of the data object in the presence of site failures. The most commonly used scheme is voting. Voting and its vari- ants are unaffected by network partitions. When network partitions cannot occur, better performance can be achieved with available copy protocols. Common measures of dependability include reliability, which is the probability that a replicated object will remain constantly available over a fixed time period. We investigate the reliability of replicated data objects managed by voting, available copy and their variants. Where possible, closed-form expressions for the reliability of the various consistency protocols are derived using standard Markovian assumptions. In other cases, numerical solutions are found and validated with simulation results. TR #88-23 Cost $4 Regeneration Protocols for Replicated Objects Darrell D. E. Long University of California, Santa Cruz Jehan-Francois Paris University of Houston Abstract The reliability and availability of replicated data can often be increased by generating new replicas when some become inaccessible due to system malfunctions. This tech- nique has been used in the Regeneration Algorithm, a replica control protocol based on file regeneration. The read and write availabilities of replicated data managed by the Regeneration Algorithm are evaluated and two new regeneration protocols are presented that overcome some of its limitations. The first protocol combines regenera- tion and the Available Copy approach to improve availability of replicated data. The second combines regeneration and the Dynamic Voting approach to guarantee data consistency in the presence of network partitions while maintaining a high availability. Expressions for the availabilities of repli- cated data managed by both protocols are derived and found to improve significantly on the availability achieved using extant consistency protocols. TR #88-07 Cost $10 The Management of Replication in a Distributed System Darrell D. E. Long (Ph.D. Thesis) University of California, San Diego Abstract The field of consistency control protocols for replicated data objects has existed for about ten years. Its birth coincides with the advent of distributed data bases and the communications technology required to support them. When data objects are replicated around a computer network, a protocol must be chosen to ensure a consistent view to an accessing process. The replicas of the data object are then said to be mutually consistent. The protocols used to insure mutual consistency are known as replica control or consistency control protocols. There are several advantages to a distributed system over a single processor system. Among these are increased computing power and the ability to tolerate partial failures due to the malfunction of individual components. The redun- dancy present in a distributed system has been the focus of much research in the area of distributed data base systems. Another benefit of this natural redundancy, along with the relatively independent failure modes of the processors, is that it allows the system to continue operation even after some of the processors have failed. This can be used to construct data objects that are robust in the face of par- tial system failures. The focus of this dissertation is the exploitation of the redundancy present in distributed systems in order to attain an increased level of fault tolerance for data objects. The use of replication as a method of increasing fault tolerance is a well-known technique. Replication introduces the additional complexity of maintaining mutual consistency among the replicas of the data object. The pro- tocols that manage the replicated data and provide the user with a single consistent view of that data are studied, and a comprehensive analysis of the fault tolerance provided by several of the most promising protocols are presented. Several techniques are employed, including Markov analysis and discrete event simulation. Simulation is used to con- firm and extend the results obtained using analytic tech- niques.