PARIS, J. F. 1986. Voting with witnesses: A consistency scheme for replicated files. In Proceedings of the IEEE International Conference on Distributed Computing Systems (ICDCS) (Cambridge, Mass.). IEEE Computer Society Press, Los Alamitos, Calif., 606--612.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....a refreshment protocol for the shares that worked both correctly and efficiently in asynchronous systems. Another problem of special interest is reducing the amount of resources required to implement a replicated service. The number of replicas can be reduced by using f replicas as witnesses [Par86, LGG 91] that are involved in the protocol only when some full replica fails. It is also possible to reduce the number of copies of the state to f 1 but the details remain to be worked out. We have shown how to implement a Byzantine fault tolerant file system. It would be interesting to use ....

J.-F. Paris. Voting with witnesses: A consistency scheme for replicated files. In Proc. of the 6th International Conference on Distributed Computer Systems, pages 606--612. IEEE, 1986.

....models assume that the failure and repair rates of components are exponentially distributed. This assumption is often made more for analytic simplicity than out of a conviction that it is the best model of reality. For example, recent studies of replicated data that employ Markov models [9, 6] depend on that assumption. The exponential hypothesis is rigorously testable, although the process of gathering the data and the problem of interpreting it are non trivial. Two of the impor Supported in part by faculty research funds and a Regents Junior Faculty Fellowship from the University ....

J.-F. Paris, "Voting with witnesses: A consistency scheme for replicated files," in Proceedings 6 th International Con- ference on Distributed Computing Systems, (Cambridge), pp. 606 612, IEEE, 1986.

....the token if required. When dirty cache contents are to be sent to a server, they are sent to the primary. The primary then asks each secondary in the view to apply the updates. Once each server has applied the updates the read write request is returned to the client. Finally, Echo uses witnesses [15] to reduce the storage requirements of some replicas. 2.2 Harp Harp is a primary copy server protocol [10] The protocol uses witnesses [15] uninterruptible power supplies, and stable RAM to lower the cost of replication. As Harp is a server protocol, there are no mechanisms to control, nor a ....

....in the view to apply the updates. Once each server has applied the updates the read write request is returned to the client. Finally, Echo uses witnesses [15] to reduce the storage requirements of some replicas. 2. 2 Harp Harp is a primary copy server protocol [10] The protocol uses witnesses [15], uninterruptible power supplies, and stable RAM to lower the cost of replication. As Harp is a server protocol, there are no mechanisms to control, nor a provision for, client caching. Primary copy protocols are the same as primary site protocols except that the file system is divided into ....

J.-F. Paris, "Voting with Witnesses: A Consistency Scheme for Replicated Files," Proc. Sixth Int'l Conf. Distributed Computing Systems, pp. 606-612, May 1986.

.... (DCS) in the presence of failures [14] 17] Numerous algorithms have been proposed for maintaining consistency among replicated data objects [6] 7] 16] 21] Included in these is a class of votingbased algorithms which tolerate both site and communication link failures [4] 8] 15] [19], 20] Recently, regeneration of failed copies of data objects has been proposed to improve availability and reliability of data [1] 3] 10] 14] 18] In [3] Adam and Tewari introduced an algorithm called RVC which utilizes a selective regeneration and recovery mechanism. A theoretical ....

....is less than 0.90. However, except for this case, regeneration should always be disabled. Since regeneration is an expensive operation, particularly for large data objects, it should be avoided whenever possible. These previous two results are also similar to a theoretical result obtained by Paris [19], where the reliability of a data object consisting of n real copies and m witnesses (witnesses are similar to virtual copies) is the same as that of a data object consisting of n m real copies. Paris s claim held for all combinations of real copies and witnesses. Adding sites with virtual copies ....

J.F. Paris, "Voting with Witnesses: A Consistency Scheme for Replicated Files," Proceedings of the 6th IEEE International Conference on Distributed Computing Systems, pages 606-612, June 1986.

....latency. Drawback (ii) is true of all algorithms which addresses network partitioning. Finally, drawback (iii) has been addressed by some enhancements to QC which allow dynamic reconfiguration of the quorums. As a replica control protocol, QC has received much attention from researchers [1, 3, 6, 7, 13, 15, 14, 19, 8, 18] . Its implementation, however, is not widespread. 3 Protocol Implementations Replica control protocols are perhaps among the most difficult to implement in a distributed database management system. The reasons are many fold: First, due to the nature of replication there is a dramatic increase ....

J. F. Paris. Voting with Witnesses: A Consistency Scheme for Replicated Files. In Proceedings of the Sixth International Conference on Distributed Computing Systems, pages 606--612, June 1986.

....has the current version number, the data can be copied from that representative, instead of from a remote node. In VWG it is unnecessary to create ghosts for weak representatives; after a crash the weak representative disappears completely. To cut down on the storage requirements witnesses [11] can be used. Witnesses are copies that only maintain the version number of the copy, and not the data themselves. In a quorum there has to be at least one real copy. Otherwise witnesses have no effect on the voting mechanism. A concept similar to a ghost is a temporary witness . A temporary ....

J.-F. Paris, "Voting with Witnesses: A Consistency Scheme for Replicated Files," Proc. of the 6th Int. Conf. on Distr. Computing Systems, pp. 606-620, Cambridge, MA, May 1986.

....control method. It was shown in [JM90] that the hybrid scheme provides better data availability than the basic dynamic voting protocol. 7.4 Voting with witnesses The replicas in the basic weighted voting scheme store a version number and the value of the data item. The voting with witnesses method [Par86] replaces some of the replicas by witnesses, which are copies that only contain the version number but no data. The witnesses are assigned votes and will cast them when they receive voting requests from transactions. They provide the transactions with their version numbers which are used to ....

....a read operation uses nodes A and C as quorum group, the version number returned by the witness will testify that the value at node A is out of date. It is thus possible for a read operation to obtain permission from a read quorum group and not find the current value of the data. An analysis in [Par86] showed that the data availability in a system with two full replicas and one witness is slightly lower than a system with three full replicas. The voting with witnesses scheme can effectively raise the data availability without having to replicate the data item. 7.5 Voting with Ghosts A system ....

[Article contains additional citation context not shown here]

J.-F. Paris. Voting with witnesses: A consistency scheme for replicated files. In Proceedings of the 6th International Conference on Distributed Computing Systems, pages 606--612. IEEE, 1986.

....accessing the resource because queueing delays can be reduced by load sharing. When data is replicated, we must maintain data consistency among the copies at different nodes. Many replica control protocols have been proposed for this end and the design goal has been to maximize data availability [Gif79, Par86, ES83]. Such protocols generally involve a relatively large fraction of nodes with replicas in the execution of a write operation. As a result, when the number of write operations is substantial, the number of requests that arrive at a particular node may not be much smaller than the requests arriving ....

J.-F. Paris. Voting with witnesses: A consistency scheme for replicated files. In Proceedings of the 6th International Conference on Distributed Computing Systems, pages 606--612. IEEE, 1986.

.... normal and failure modes [3] which have different values for read and write quorums) The system can also change the quorum assignment in the schemes presented in [4, 5, 6] and the vote assignment can be changed in the scheme described in [7] Other protocols based on voting are presented in [8, 9, 10]. The problem of assigning votes to achieve mutual exclusion is addressed by Garcia Molina and Barbara in [11] When the quorum for each operation is a majority of all votes assigned, each operation will have mutually exclusive access to the data. In general, mutual exclusion can be guaranteed by ....

J.-F. Paris, "Voting with witnesses: A consistency scheme for replicated files," in Proceedings of the 6th International Conference on Distributed Computing Systems, pp. 606--612, IEEE, 1986.

....Theory (ICDT 90) Paris, France, Dec. 1990. Lecture Notes in Computer Science 470. Berlin, Heidelberg, New York: Springer Verlag, pages 318 332 y e mail: borghoff lan.informatik.tu muenchen.dbp.de Voting Strategy Transparent access to replicated files has been extensively investigated [3, 4, 8, 10, 12, 13, 16, 17, 19, 20, 28, 29, 31, 35, 36, 39, 40, 41, 46]. The replication method proposed here is a modification of a file replication method due to Thomas [45] and Gifford [24] Our method encompasses replication methods that allow for dynamic changes of the degree of replication as well as for dynamic changes of the location of replicas [21, 26, 27, ....

J.-F. Paris. Voting with witnesses: A consistency scheme for replicated files. In Proc. 6th IEEE Int. Conf. on Distributed Computing Systems, pages 606--620, Cambridge, MA, May 1986. Los Alamitos, CA: IEEE Comp. Soc. Press.

....can issue a message to a group of servers in parallel. The network delays are considered negligible compared with disk access delays which dominate service demands at the servers. Also, we assume that only nodes can fail, and the network is reliable. In modeling node failures, we follow [9] and [13] in assuming that a node participating in the operation cannot repair or fail during the operation execution (operations are instantaneous with regard to failures and repairs) In other words, when a server is first contacted by an operation, it can be found either up or down (with the ....

....very effective propagation of missing updates by providing stale replicas with a strong hint about the location of current replicas. Finally, our approach blends well with other ideas in replica control like dynamically defined quorums (see, e.g. 9] and protocols with witnesses proposed in [13]. Witnesses are the nodes that keep state information but no actual data. They participate in read and write quorums, update their version numbers, etc. but never apply actual operations. The idea is to cut disk space overhead without compromising availability of the system. One can incorporate ....

J.-F. Paris. Voting with witnesses: A consistency scheme for replicated files. In Proc. of the IEEE Int. Conf. on Distr. Comp, pp. 606-621, 1986.

....physically multiple copies of the file exist, nevertheless, they must behave logically as if there is a single copy in the system. Many replica control protocols have been developed to synchronize read and write accesses to replicated files so that the replicas behave logically as a single copy [11, 12, 13, 14, 15, 16]. Read and write operations that access replicated files are synchronized by obtaining permissions from nodes that constitute a read and write quorum group, respectively, and the read and write quorum groups are subsets of nodes of the residence set R [7] consisting of the nodes that maintain a ....

J.-F. Paris, "Voting with witnesses: A consistency scheme for replicated files," in Proceedings of the 6th International Conference on Distributed Computing Systems, pp. 606--612, IEEE, 1986.

....processing latency. Drawback (ii) is true of all algorithms which addresses network partitioning. Finally, drawback (iii) has been addressed by some enhancements to QC which allow dynamic reconfiguration of the quorums. As a replica control protocol, QC has received much attention from researchers [1, 9, 16, 18, 19, 29, 28, 27, 36, 20, 35] . Its implementation, however, is not widespread. The Majority Voting protocol is perhaps the version of the protocol that is typically considered. 4 Simulation Model We made use of a simulation testbed to investigate the performance of the PC and QC protocols in a client server environment. The ....

....As has already been mentioned, the PC protocol implemented on the Harp distributed file system makes use of special hardware to reduce message and logging overhead. Another example is that under QC there are schemes for using bystanders (or witnesses) in lieu of data copies to reduce overhead [27, 28]. Also, under both PC and QC, the copies can be organized in a multi level hierarchy [20] to reduce the amount of message overhead (however, such a technique is not applicable to our model where the number of copies is small. The key point is that there are many techniques which can be explored ....

J. F. Paris. Voting with Witnesses: A Consistency Scheme for Replicated Files. In Proceedings of the Sixth International Conference on Distributed Computing Systems, pages 606--612, June 1986.

....and reliability models assume that the failure and repair rates of components are exponentially distributed. This assumption is often made more for analytic simplicity than out of a conviction that it is the best model of reality. Recent studies of replicated data that employ Markov models [12, 8] depend on that assumption. The exponential hypothesis is rigorously testable, although the process of gathering the data and the problem of interpreting it are non trivial. Two of the important statistics that are derived are mean time to failure (MTTF) and mean time to repair (MTTR) MTTF is ....

J.-F. Paris, "Voting with witnesses: A consistency scheme for replicated files," in Proceedings 6 th International Conference on Distributed Computing Systems, (Cambridge), pp. 606--612, IEEE, 1986.

....nodes remains connected is an NP hard problem. Most performance analyses of partition resilient schemes (such as static and dynamic voting) assume that the communication links are infallible, thereby excluding the possibility of partitioning due to link failures precisely the event of interest [2 7]. Other approaches assume that all the link failure probabilities are identical and deal with very simple, small networks [3, 6, 8] Enumeration of all failure states of the network to find the partition probabilities has also been used [9] but this is very impractical for large networks. 10] ....

....availability of the system. However, while the read throughput of the system does go up with the number of copies, there is a copy synchronizing overhead associated with the write transactions. Voting algorithms have been designed to maintain multiple copies consistently in a replicated database [2, 6, 7, 18 28]. The main idea is to dis allow concurrent updating of the same data item by two or more transactions. For instance, if the network is partitioned into smaller sub networks, the voting algorithm should be so designed as to allow updates in only one of the sub networks. In the following, we ....

J.-F. Paris, "Voting with Witnesses: A Consistency Scheme for Replicated Files," in Proceedings of the 6th International Conference on Distributed Computing Systems, (Massachusetts), pp. 606--612, May 1986.

....of these nodes then have the version number as one more than the highest version number. The quorum conditions ensure that a node with the highest of the version numbers in a read write quorum has the latest copy of the data. Many further improvements of voting algorithm have also been suggested [2, 3, 4, 5, 8, 10, 11, 16, 17, 18, 19]. The voting scheme has a high communication cost, a request for an operation is typically send to all the sites in the system and a group of sites constituting a majority of votes satisfies the quorum. Some extensions of the voting algorithm have been proposed to reduce the communication costs ....

J.F.Paris, "Voting with witnesses: A consistency scheme for replicated files", in proc. of sixth IEEE int. conf. on dist. computing systems, pp. 606612, June 1986.

....the majority quorum with respect to the current update cardinality and an update is performed on replicas at all operational nodes. The dynamic voting method can thus allow the system to adapt its quorum requirement to changes in the system state. Voting with witnesses algorithms is discussed in [Par86, Par94]. The replicas in the basic weighted voting method scheme store a version number and the value of the data item. The voting with witnesses method replaces some of the replicas by witnesses which are copies that contain only the version number but no data. The witnesses are assigned votes and will ....

Paris, J., Voting with Witnesses: A Consistency Scheme for Replicated Files, Proceedings Sixth Intl. Conf. Distributed Computing Systems, 1986, p.606-612.

....conflicting updates. As a result, they provide lower data availabilities than the other protocols. The best known quorum oriented protocols include majority consensus voting (MCV) weighted voting (WV) 5] dynamic voting (DV) 4] dynamic linear voting (DLV) 9] and voting with witnesses (VWW) [15]. A common feature of all replication control protocols is the use of metadata to record the states of the replicas. These metadata nearly always include a version number, that is an integer that is incremented each time the replicated data are modified. Protocols such as optimistic available ....

J.-F. Paris, "Voting with Witnesses: A Consistency Scheme for Replicated Files," Proc. 6th International Conference on Distributed Computing Systems, (1986), pp. 606--612.

....protocols. Pessimistic protocols require at least three replicas to implement a robust consensus and improve upon the availability of unreplicated data. However one of these replicas can be replaced by a much smaller witness without significantly affecting the availability of the replicated data [16]. Witnesses are very small entities that hold no data but maintain enough information to identify the replicas that contain what it believes to be the most recent version of the data. Conceptually this information could be a timestamp containing the time of the latest update. Since it is quite ....

Paris, J.-F.: Voting with witnesses: a consistency scheme for replicated files. Proc. 6th Int. Conf. on Distributed Computing Systems, (1986) pp. 606--612.

....of voting protocol is the fact these protocols require at least three voting entities to improve upon the availability afforded by unreplicated data. Fortunately, one of these three entities can be a witness, that is an entity containing the same metadata as a regular replica but no data [12]. Cohort sets make witnesses especially attractive because they reduce to an absolute minimum the storage costs and the access costs. To evaluate the storage cost of a witness we need to distinguish between static voting protocols where witnesses do not need to be included in cohort sets and ....

J.-F. Paris, "Voting with Witnesses: A Consistency Scheme for Replicated Files," Proc. 6th Int. Conf. on Distributed Computing Systems, (1986), pp. 606--612. 135

....rate . The system is assumed to exist in statistical equilibrium and to be characterized by a discretestate Markov process. These assumptions are required for a steady state analysis to be tractable and, in fact, have been made in most probabilistic studies of the availability of replicated data [2, 8, 12 13]. Combinational models that do not require any assumptions about failure and repair distributions have been proposed [15 16] but these models cannot distinguish between available states and recovery states. The availability analysis of a replicated object subject to network partitions is ....

J.-F. Paris, "Voting with Witnesses: A Consistency Scheme for Replicated Files," Proc. 6th ICDCS , (1986), pp. 606-612.

No context found.

PARIS, J. F. 1986. Voting with witnesses: A consistency scheme for replicated files. In Proceedings of the IEEE International Conference on Distributed Computing Systems (ICDCS) (Cambridge, Mass.). IEEE Computer Society Press, Los Alamitos, Calif., 606--612.

No context found.

J.-F. Paris. Voting with witnesses:: A consistency scheme for replicated files. In Proceedings of the 6th International Conference on Distributed Computing Systems (ICDCS), pages 606--612, Washington, DC, 1986. IEEE Computer Society.

No context found.

J.-F. Paris. Voting with witnesses:: A consistency scheme for replicated les. In Proceedings of the 6th International Conference on Distributed Computing Systems (ICDCS), pages 606-612, Washington, DC, 1986. IEEE Computer Society.

No context found.

J.F. Paris. Voting with Witnesses: A Consistency Scheme for Replicated Files. In Proceedings of the International Conference on Distributed Computing Systems, pages 606--612, May 1986.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC