H. Garcia and D. Barbara. "How to assign Votes in a Distributed System". J. ACM, May 1985.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....and version number associated with the variable. Clients expect that query returns the value with the largest version number over all that have ever been written by update. Although such a service could be implemented using the state machine approach, a cheaper approach called a quorum system [45, 105, 41] su#ces. In a quorum system, servers are organized into sets called quorums, where every pair of quorums intersect. 2 And, instead of requiring that all non faulty servers receive and process the same requests (as in the state machine approach) a quorum system requires only that, for each ....

H. Garcia-Molina and D. Barbara. How to assign votes in a distributed system. Journal of the ACM, 32(4):841--860, October 1985.

....stored at the majority of the processors as the correct data. The work of this flavor can be found in classical problems of agreement and consensus [AW98, LS 82, DP 88] system level diagnosis [S86, P96a, DP 96] distributed database management [DG 85, H84] quorum systems [G79, GB85, SB94, PW95, W96] and fault local mending [KP95a, KP95b] To give a concrete example, suppose that all processors in a distributed network collectively store some value and suppose that this value is distorted in some of the processors (distortions could be due to various reasons, even due to ....

H. Garcia-Molina, D. Barbara, How to assign votes in a distributed system, J. of ACM, vol 5, 32:4(841-860), 1985.

....enforces the modification for each new configuration. The second one, which has been selected here, considers the modification decision as a cooperative decision. As a consequence, the opportunity of changing is proposed to the current set of cooperating agents. The cooperating agents vote [8] to accept or to deny the configuration change. The vote policy is specified by the cooperative application. According to the vote result, the valid configuration is changed or not, as shown in Figure 3. The current cooperation structure remains unchanged if the modification is refused. Vote has ....

....According to the vote result, the valid configuration is changed or not, as shown in Figure 3. The current cooperation structure remains unchanged if the modification is refused. Vote has been retained as a generic mechanism to implement a group decision technic. Moreover, combined with weights [8], votes offer very rich policies according to the different positions and importance of the cooperative group members. Some classical examples of votes policies are unanimity, majority vote, but more complex policies can be obtained such as giving the whole decision to a subset of agents or to a ....

H. Garcia-Molina and D. Barbara. How to assign votes in a distributed system. Journal of the ACM, 32(4):841-860. October 1985.

....using probabilistic quorum constructions can be driven further toward zero when updates are suciently dispersed in time, making probabilistic quorum constructions useful in a wider variety of settings. 1.2. Related work Strict quorum systems have been extensively studied and measured (cf. [17, 40, 26, 16, 20, 5, 14, 10, 1, 34, 37, 35]) Byzantine quorum systems were introduced in [29] and further studied in [31, 8] Because of the possibility of inconsistency admitted by probabilistic quorum systems, they are most attractive for systems in which some level of inconsistency can be tolerated, and in particular, where the ....

H. Garcia-Molina and D. Barbara. How to assign votes in a distributed system. Journal of the ACM, 32(4):841-860, October 1985.

....complexity almost halves compared with the one by Raymond, but it is still# n) where n is the number of processes in the distributed system. Our algorithm is a permission based one and uses the concept of k coterie, which we will introduce in section 2 as an extension of the concept of coterie[2]. Informally, a k coterie is a set of groups of processes, called quorums, such that any (k 1) quorums contain a pair of quorums intersecting each other. In our algorithm, a process can enter a critical section, if and only if it receives permission from every process in a quorum. Since no (k ....

H. Garcia-Molina and D. Barbara. How to assign votes in a distributed system. Journal of the ACM, 32(4):841--860, October 1985.

....given one vote each and jU j = n and n is odd, then the collection of all subsets of U of size (n 1) 2 is a quorum system which is also a non dominated coterie. One of the first algorithms for data replication that uses quorum systems defined on weighted voting is described in [17] However in [16], Garcia Molina and Barbara show that not all quorum assignments can be characterized by weighted voting systems. Alternative methods of constructing quorum systems and coteries have been widely researched. Some such schemes are the Finite Projective Plane of Maekawa [27] the tree system of [2] ....

H. Garcia-Molina and D. Babara. How to assign votes in a distributed system. Journal of the ACM, 32(4):841--860, 1985.

....are allowed with B, by the quorum fa; cg. Hence B is better than A. For another example, if A is a coterie ffa; b; cgg and B is a coterie ffa; bgg, then B is better than A, since an operation in B needs only to access two sites a; b, but an operation in A needs to access three sites a; b; c. In [9], A is said to be dominated by B in both of the above cases. More formally, we have the following definition. ffl Let R, S be coteries on G, R dominates S iff R 6= S and, for each H 2 S, there is an H 0 2 R such that H 0 H . A coterie S on G is dominated iff there is another coterie on G ....

....5.2 Non domination We have seen that non domination is a desirable property of a protocol. Here we shall show that if we choose to use ND coterie or wr coteries in the hypercube quorum protocol, the resulting protocol is also ND. We apply the following theorem in our proof. Theorem 2. 1 of [9]: Let S be a coterie under U . Coterie S is dominated iff there exists a group G U such that 1. G is not a superset of any group in S. 2. G has the intersection property. That is, for all H 2 S, G H 6= OE. This Theorem can be generalized for the wr coteries as follows: Theorem 3 Let A = ....

[Article contains additional citation context not shown here]

H. Garcia-Molina, D. Barbara, How to Assign Votes in a Distributed System, Journal of the Association for Computing Machinery, Vol.32, No.4, 1985, pages 841-860.

.... been discussed in [3, 4, 1] more recent results can be found in [13] An encyclopedic exposition of the applications of the transversal hypergraph problem can be found in [6, 7] We brie y state from there certain problems in the design of relational databases [15, 16] in distributed databases [8], and in model based diagnosis [5] Another interesting connection was pointed out between the transversal hypergraph problem and the rapidly growing eld of knowledge discovery in databases, or data mining [9, 17] In this paper we present and experimentally evaluate a heuristic algorithm for ....

H. Garcia-Molina and D. Barbara. How to assign votes in a distributed system. Journal of ACM, 32(4):841-860, 1985.

....over the network. Recent developments in quorum consensus are mainly focused on 1) minimizing the total communication costs for processing a given set of transactions, and 2) minimizing the number of remote sites to be communicated while assembling a quorum. A number of quorum consensus protocols [5, 9, 11, 12, 13, 14, 16, 17, 20, 21] have been developed for these purposes. Note that in quorum consensus, since messages are sent (possibly by the multicast mechanism [19] to the multiple nodes in a quorum in order to ensure consistency of the operations the delays by passing messages through a long distance communication ....

....Preliminaries A network is represented by a weighted graph G = V; E) where for each edge (physical link) u; v) 2 E, d u;v denotes the communication delay to send a unit message along the link from node u to v. A set S of subsets of V is coterie for G if and only if the following conditions hold [9, 20]: 1. Intersection: 8Q 1 ; Q 2 2 S, Q 1 Q 2 6= 2. Non redundancy: 8Q 1 ; Q 2 2 S, Q 1 6 Q 2 . 3. Connectivity: 8Q 2 S, the subgraph GQ induced [7] by Q from G is connected. Each element Q in a coterie is called a quorum. The intersection property guarantees that any pair of quorums in a ....

H. Garcia-Molina and D. Barbara, How to Assign Votes in a Distributed Systems, J. ACM, 32(4), 841-860, 1985.

....to p i , then it is allowed to access the resource. To ensure mutual exclusion the sets S are required to satisfy the intersection property: For any i and j, S i S j # f. These and related concepts were formalized and analyzed in terms of the notions of quorums and coteries [8] 13] [7]. In (k 1) exclusion, up to k processes are allowed to access the resource simultaneously. Thus, if we consider k 1 sets of sites that grant permission to access the resource then there must exist at least two among these k 1 sets with a nonempty intersection. The (k 1) exclusion problem ....

H. Garcia-Molina and D. Barbara, "How to Assign Votes in a Distributed System," J. ACM, vol. 32, no. 4, pp. 841--860, Oct. 1985.

....basic algorithmic techniques are very efficient and they render the algorithm suitable for an effective implementation. Using majorities is a special case of quorum systems [6] A simple quorum system (also called coterie) is a collection of sets such that any two sets, called quorums, intersect [5]. A more refined approach divides the quorum system into a collection of read quorums and a collection of write quorums such that any read quorum intersects any write quorum. Such systems have been used to implement distributed mutual exclusion [5] and data replication protocols [4, 7] Quorums ....

....such that any two sets, called quorums, intersect [5] A more refined approach divides the quorum system into a collection of read quorums and a collection of write quorums such that any read quorum intersects any write quorum. Such systems have been used to implement distributed mutual exclusion [5] and data replication protocols [4, 7] Quorums can be used with replicated data in transaction style synchronization that limits concurrency (cf. 2] whereas our goal and the goal of [1] is to reduce restrictions on asynchrony and concurrency. In this paper we present a service that emulates ....

H. Garcia-Molina and D. Barbara, "How to Assign Votes in a Distributed System," J. of the ACM, vol. 32, no. 4, pp. 841-860, 1985.

....H, and (b) no proper subset of T does. A hypergraph is simple if no hyperedge is totally contained in another one. transversal hypergraph is the problem of generating tr(H) given a simple hypergraph H, and is a common subproblem in many applications, including databases [11] distributed systems [9], etc. The precise complexity of this problem is still unknown. However, it was shown in [8,12] that it can be solved in time O(c log c ) where c is the combined size of the input and the output. Recently, a heuristic algorithm with good performance in practice was presented in [16] The main ....

H. Garcia-Molina and D. Barbara. How to assign votes in a distributed system. Journal of ACM, 32(4):841-860, 1985.

....serialization order can be The earlier version of this paper was published in CATS 96 1 ensured. Thus, an appropriate management of replicated data involves a compromise between two conflicting goals: maximizing data availability and maintaining consistency of data. Quorum consensus (QC) methods [5,6] are frequently used in managing replicated data. Using a QC method, an operation of a transaction issued at a site can proceed only if permission is granted by a group of other sites storing the replicas of the manipulating data. A general protocol of a QC method for processing a transaction ....

H. Garcia-Molina and D. Barbara, How to Assign Votes in a Distributed Systems, J. ACM, 32(4), 841-860, 1985.

....it can be solved by an efficient polynomial time algorithm. Key words: concurrency control, replicated data management, optimization, quorum consensus method. 1 Introduction The problem of managing replicated copies of data in a distributed database has received a great deal of attention [4,5,6,9,11,15] throughout the last decade. The main issue is to provide high data availability through data replication. Meanwhile, the replicated copies of data must be kept mutually consistent by synchronizing transactions at different sites so that a global serialization order can be ensured. To pursue ....

....to provide high data availability through data replication. Meanwhile, the replicated copies of data must be kept mutually consistent by synchronizing transactions at different sites so that a global serialization order can be ensured. To pursue mutual consistency, a quorum consensus (QC) method [2,4,5,12] has been proposed for managing replicated data. In a QC method, an operation of a transaction issued at a site in a distributed database system can proceed only if permission is granted by a group of other sites storing the replicas of the data. A basic QC method [2,4,11] can be described as ....

[Article contains additional citation context not shown here]

H. Garcia-Molina and D. Barbara, How to Assign Votes in a Distributed Systems, J. ACM, 32(4), 841-860, 1985.

....protocols for the latter service that tolerate faulty clients in addition to faulty servers in Section 6. We conclude in Section 7. 2 Related work Our work was influenced by the substantial body of literature on quorum systems for benign failures and applications that make use of them, e.g. [Gif79, Tho79, Mae85, GB85, Her86, ET89, CAA90, AE91, NW94, PW95]. In particular, our grid construction of Section 4 was influenced by grid like constructions for benign failures (e.g. CAA90] and we borrow our definitions of domination and load from [GB85] and [NW94] respectively. Quorum systems have been previously employed in the implementation of ....

.... applications that make use of them, e.g. Gif79, Tho79, Mae85, GB85, Her86, ET89, CAA90, AE91, NW94, PW95] In particular, our grid construction of Section 4 was influenced by grid like constructions for benign failures (e.g. CAA90] and we borrow our definitions of domination and load from [GB85] and [NW94] respectively. Quorum systems have been previously employed in the implementation of security mechanisms. Naor and Wool [NW96] described methods to construct an access control service using quorums. Their constructions use cryptographic techniques to ensure that out of date (but ....

H. Garcia-Molina and D. Barbara. How to assign votes in a distributed system. Journal of the ACM 32(4):841--860, October 1985.

....to be mutually consistent. In the famous two phase locking (2PL) protocol [7, 5] one of the replicas is locked for read and all the replicas are locked for write, i.e. read one write all principal is used. The method is not efficient for write dominated applications. In the quorumbased method [1], some numbers Q r and Qw of the replicas are locked for read and write, respectively. Here, Q r Qw is larger than the total number a of the replicas. Q r and Qw are decided based on how frequently read and write are issued. The distributed applications are modeled to be a collection of multiple ....

H. Garcia-Molina and D. Barbara, "How to Assign Votes in a Distributed System," Journal of ACM, Vol 32, No.4, pp. 841-860, 1985.

....may still be possible for one part of the network to contain a complete quorum. Of course, it is quite impossible for two disconnected parts of the system to both contain complete quorums ) Considerable attention is given in the literature to a special type of quorum system called a coterie (see [GB85] and [IK90] A coterie is a quorum system in which the quorums are not allowed to fully contain each other. A subclass of special interest is that of nondominated coteries (or NDCs) which are better than other coteries in terms of fault tolerance and communication cost. This subclass is defined ....

....This subclass is defined as follows. Given two coteries S 1 and S 2 over the same universe U , we say that S 2 dominates S 1 if S 2 #= S 1 and for every quorum S #S 1 there is a quorum T #S 2 such that T # S. An NDC is a coterie which is not dominated by any other coterie (see [GB85]) 1.2. Load balancing. There are many types of quorum systems, and many parameters of quorum systems a#ecting the applications using them. Such parameters include quorum sizes (a#ecting communication costs) and the number of quorums LOAD BALANCING IN QUORUM SYSTEMS 225 (a#ecting immunity to ....

[Article contains additional citation context not shown here]

H. Garcia-Molina and D. Barbara, How to assign votes in a distributed system,J. Assoc. Comput. Mach., 32 (1985), pp. 841--860.

....of the object have to be mutually consistent. The two phase locking (2PL) protocol [1,2] locks one of the replicas for read and all the replicas for write. The 2PL protocol is not efficient for write dominated applications because all the replicas are locked for write. In the quorum based protocol [3], some numbers Q r and Qw of the replicas named quorum numbers are locked for read and write, respectively. Here, a constraint Q r Qw a for the number a of the replicas has to be satisfied. Distributed applications are modeled to be a collection of multiple objects which are cooperating. ....

.... by an object o conflict if the result obtained by applying op 1 and op 2 to o depends on the computation order of op 1 and op 2 [1] In this paper, we propose a novel locking scheme for replicated objects named OBL (object based locking) protocol, which is an extension of the quorum based protocol [3] to the replicas of the abstract objects. Before computing a method op t on an object o, some quorum number Q t of the replicas of o are locked in the abstract mode for op t . Q t depends on how frequently op t is invoked. The more frequently op t is invoked, the smaller Q t gets. Suppose a pair ....

[Article contains additional citation context not shown here]

Garcia-Molina, H. and Barbara, D., "How to Assign Votes in a Distributed System," Journal of ACM, Vol 32, No.4, 1985, pp. 841-860.

....35 efficient, and convergent optimal) is unlikely to exist for general networks. Works on Quorum Consensus (such as [Agrawal and Bernstein 1991; Gifford 1979; Kumar 1991; Thomas 1979; Triantafillou and Taylor 1991] Voting and Coterie (such as [Agrawal and El Abbadi 1990; Adam and Tewari 1993; Garcia Molina and Barbara 1985; Herlihy 1987; Jajodia and Mutchler 1990; Paris 1986; Spasojevic and Berman 1994] refer to performance in the presence of failures. They address the issue of how to dynamically adjust the read write quorums and votes in order to minimize the data accesses in case of site failures and network ....

Garcia-Molina, H. and Barbara, D. 1985. How to assign votes in a distributed system. J. ACM 32, 4, 841--860.

....critical resource by competing sites. The main drawback of permission based mutual exclusion algorithms is that the communication cost to enter critical section is directly proportional to the size of quorums. Much research has been done to optimize communication and minimize the size of quorums [Mae85, GB85, AE91, Kum91, CAA92]. Maekawa [Mae85] has shown that under certain conditions (which are strongly desirable in a distributed system) the optimal quorum size is p N , where N is the total number of sites in the network. This optimal size is achievable when the sites in the network can be organized logically as a ....

....mutual exclusion, any S i must have at least one site in common with any other set. This can be stated formally as follows: The Intersection Property. For any two sets S i and S j , S i S j 6= OE. These concepts have been formalized in terms of the notions of quorums [Gif79] and coteries [GB85]. Maekawa [Mae85] has shown that within a constant factor, the quorum size q is optimal when q = p N . Furthermore, in the case of a finite projective plane of N points, each site is assigned a unique quorum S i , and the conditions of intersection and optimality are also satisfied. Such a plane ....

[Article contains additional citation context not shown here]

H. Garcia-Molina and D. Barbara. How to assign votes in a distributed system. Journal of the Association of the Computing Machinery, 32(4):841--860, October 1985.

.... account balances may change because other transactions may have affected the balance) and unacceptable for others (an item is out of stock and cannot be sold) This technique does a good job in solving a fundamental issue: standard ways of propagating updates to replicas (eager replication [26] [25], lazy replication [21] 56] make deadlocks increase as the cube of the number of sites and as the fourth power of transaction size, thereby rendering the systems unscalable. On the other hand, the two tier technique may result in an unacceptable number of failed transactions (after the clients ....

# H. Garcia-Molina and D. Barbar, "How to Assign Votes in A Distributed System," J. ACM, vol. 34, no. 4, pp. 841-860, Oct. 1985.

No context found.

H. Garcia and D. Barbara. "How to assign Votes in a Distributed System". J. ACM, May 1985.

No context found.

H. Garcia and D. Barbara, "How to Assign Votes in a Distributed System," Journal of ACM, May 1985.

No context found.

) Garcia-Molina, H. and Barbara, D.: How to Assign Votes in a Distributed System, Journal of the ACM , Vol. 32, No. 4, pp. 841--860 (1985).

No context found.

) Garcia-Molina, H. and Barbara, D., "How to Assign Votes in a Distributed System," Journal of ACM, Vol. 32, No. 4, pp. 841-860 (1985). Vol. 0 No. 0 Version Vector for Maintaining Distributed Replicas 9

First 50 documents  Next 50

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