O. Wolfson and A. Milo, "The multicast policy and its relationship to replicated data placement," ACM Trans. Database Syst., vol. 16, no. 1, pp. 181--205, Mar. 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....are connected to the backbone via direct links. The major requirement is that the total access cost for the users is within a specified bound, whereas the construction cost of the backbone network should be minimized. Similar problems arise in the efficient maintenance of distributed databases [28, 6, 5, 20, 9]. Other applications of BCCMED include location theory and manufacturing logistics (see [25, 27] and the references cited therein) In this paper, we study the complexity and approximability of BCCMED. The paper is organized as follows. In Section 2 we formally define the problem under study and ....

O. Wolfson and A. Milo. The multicast policy and its relationship to replicated data placement. ACM Transactions on Database Systems, 16(1):181--205, March 1991. Received

....model. The major di erence is that they only consider read requests. They present a constant factor approximation algorithm that aims to minimize the total cost. This algorithm is based on solving a linear programming relaxation of the problem and rounding the obtained solution. Milo and Wolfson [15] show that the static data management problem in arbitrary networks is already NP hard in the total communication load model. Further, they present polynomial time algorithms for completely connected networks, trees, and rings that calculate optimal placements in the total communication load ....

O. Wolfson and A. Milo. The multicast policy and its relationship to replicated data placement. ACM Transactions on Data Base Systems (TODS), 16(1):181-205, 1991.

....all but one block contain exactly B nodes increases the blocknumber by at most 2. 1. INTRODUCTION Data replication allows better performance with increased availability of data. In static environments, data replication does not have to deal with maintaining consistency among multiple copies [1, 3, 6, 11, 13, 14]. Static environments dealing with large data sets arise frequently. They include environments in which updates happen at certain, chosen points in time and in bulk, and queries are not present at the time of updating. In this paper we show how to e ectively use data replication to improve ....

O. Wolfson and A. Milo. The multicast policy and its relationship to replicated date placement. ACM Transactions on Database Systems, 16(1):181-205, Mar. 1991.

....have been extensively studied in the database community [20] Wolfson, Jajodia, and Huang [21] proposed an adaptive, le replication scheme that migrates redundant copies of les to locations where the read write activ2 ity is highest on a tree network. In an earlier study, Wolfson and Milo [22] showed that nding an optimal replication scheme, with minimal cost for a given read write pattern over general network topologies, is NP complete. Input output characterization. There are a number of studies that present models of physical disk [23, 24, 25] and disk array [17] access behavior. ....

O. Wolfson and A. Milo, \The Multicast Policy and its Relationship to Replicated Data Placement," ACM Transactions on Database Systems, vol. 16, pp. 181-205, Mar. 1991.

....to the backbone via direct links. The major requirement is that the total access cost for the users be within a specified bound, whereas the construction cost of the backbone network should be minimized. Problems of similar nature arise in the efficient maintenance of distributed databases [3, 4, 7, 15, 22]. Other applications of the Budget Constrained Connected Median Problem studied in this paper include location theory and manufacturing logistics (see [20, 21] and the references cited therein) The above problems can be cast in a graph theoretic framework as follows: Given an undirected graph G ....

O. Wolfson and A. Milo, The multicast policy and its relationship to replicated data placement, ACM Transactions on Database Systems 16 (1991), no. 1, 181--205.

....have to update all copies of an object, whereas read accesses can be satisfied by any one copy. However, none of the described models assumes that copies are updated within a multicast tree. Milo and Wolfson investigate static placement in a model that permits multicast trees for copy updates [85]. However, the multicast trees are allowed to branch out only on nodes holding a copy of the respective object, as opposed to our model which allows branchings at every node. Milo and Wolfson introduce placement algorithms that minimize the total communication load for rings, trees, and completely ....

....simultaneously. Obviously, this yields minimum congestion as well as minimum total communication load. We describe an efficient algorithm that calculates this placement. For networks with point to point connections this algorithm is more or less equivalent to the algorithm of Wolfson and Milo [85], for which they were only able to prove optimality in the total load model. The sequential running time of the algorithm is O(n) for each object, where n denotes the size of the network. Moreover, the placement can be computed efficiently in a distributed fashion by the processors of the ....

[Article contains additional citation context not shown here]

O. Wolfson and A. Milo. The multicast policy and its relationship to replicated data placement. ACM Transactions on Data Base Systems (TODS), 16(1):181--205, March, 1991.

.... of this tree [2, 3] Other applications where multicasting can improve performance are: finding idle computers in a local area network [4] implementing reliable and ordered group communication [5] sending multi media messages to a large number of subscribers [6] and accessing replicated data [7]. In this paper, we study a generalization of multicasting where a message needs to be delivered only to a subset of the multicast destinations. This type of communication requirement arises in a number of distributed applications. An example where processes need to communicate with a subset of ....

.... 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 copy of the file. A read and a write quorum group, and two different write quorum groups, must intersect for the set of copies to remain in synchronization. Clearly, collecting a read and a write quorum require communicating with a subset of the predefined ....

O. Wolfson and A. Milo, "The multicast policy and its relationship to replicated data placement," ACM Transactions on Database Systems, vol. 16, pp. 181-- 205, March 1991.

....that have not been published [9] Revised draft submitted to IEEE Transactions on Software Engineering. June 1993. Multicast has been employed to find idle computers [26] to resolve names [7] to implement ordered and causally ordered group communication [8, 4, 17, 6] to update replicated data [23], and to distribute voice and data throughout wide area internetworks [12] Only some applications of multicast benefit from flow control. In Cheriton s multicast based name service [7] for example, the originator need only capture one of the recipients responses, the others are redundant. It ....

Amir Milo and Ouri Wolfson. The multicast policy and its relationship to replicated data placement. ACM Transactions on Database Systems, 15(1):181--206, March 1991.

....database is an important and extensively studied problem in computer science. See Awerbach et al. 1992] Awerbach et al. 1993] Lund et al. 1994] Dowdy and Foster [1982] Milo and Wolfson [1988] Kumar and Segev [1993] and the references therein. As an example, consider the problem posed in Wolfson and Milo [1991] on the design of distributed databases: given a set of sites in a network we wish to select a subset of the sites at which to place copies of the database. The major requirement is that each site should be Received August 1996. 2 MARATHE, RAVI AND SUNDARAM able to access a copy of the database ....

Wolfson, O. and Milo, A. 1991. The multicast policy and its relationship to replicated data placement. ACM Transactions on Database Systems 16, 181--205.

....gives the total bandwidth consumption, but it does not correspond to the message delay. Another possible metric for the communication cost for a node s to reach a quorum is the total weight of a weighted minimum spanning tree whose root is at node s and which spans each node in the quorum (see [26]) Given a coterie, each node s chooses a quorum such that the virtual distance D s to the furthest node in the quorum is minimal. The maximal value of D s among all s will be an upper bound on the delay to reach a quorum for any node. The average value of D s of all s will give us the average ....

O. Wolfson and A. Milo. The multicast policy and its relationship to replicated data placement. ACM Transactions on Database Systems, 16(1):181--205, 1991.

....specified by the mapping q: x V R where is the set of all fragments and q(f k ,v i ) is the amount of data (in bytes) of fragment f k used at site v i to satisfy all queries issued at site v i . In this paper it is assumed that update propagation is done via a Minimal Spanning Tree (MST) [14]. In the MST strategy, a MST exists between copies of each fragment. When an update occurs at one of these copies it is propagated via the MST to all other copies. This is defined for each fragment by the MST update propagation scheme mapping m: x V x V 0,1 where: 1 if the link between ....

....If other points are allowed to act as nonparticipating intermediate points, a spanning tree of shorter total length may be possible. Finding such a tree is called the Minimum Steiner Tree problem [8] 7] 1] Orlowski in [8] presents algorithms for finding update propagation paths in such cases. In [14] though, it was stated that a nonparticipant does not send an incoming replica to two or more of its neighbours. Such use of nonparticipant points is not an issue though if the network under consideration is a fully connected and strict metric. Here, the direct link between any two points is ....

Wolfson O., Milo A., "The Multicast Policy and Its Relationship to Replicated Data Placement", ACM Transactions on Database Systems, Vol. 16, No.1, March 1991, pp.181-205

....to be equal to the cost of a minimum spanning tree of a certain graph; this graph is the subgraph of the complete graph that corresponds to the tree network restricted to the the set of nodes that have a copy of the object. The model we use in this paper is based on the model of Wolfson and Milo [1]. Further, we assume that having a copy of the object at a node has an associated storage cost that depends on the object and that node. The cost of a residence set is equal to the sum of its total read, write, and storage costs. We are interested in finding a residence set of minimum cost. We ....

....residence set problem for trees with read, write, and storage costs. The optimal residence set (file allocation) problem has been studied extensively in the literature. Dowdy and Fos ter [2] survey a number of mixed linear programming models for the file allocation problem. Wolfson and Milo [1] consider the optimal residence set problem without stor age costs for various interconnection networks (completelyconnected, tree, and ring networks) They show that the optimal residence set problem without storage costs is NPhard for general topologies, and they provide efficient al gorithms ....

[Article contains additional citation context not shown here]

O. Wolfson and A. Milo, "The multicast policy and its rela- tionship to replicated data placement," ACM Transactions on Database Systems, vol. 16, no. 1, pp. 181-205, 1991.

....cost objective function reduces the load on the communication network and the processors CPU cost in processing the messages. The problem of finding an optimal replication scheme, i.e. a replication scheme that has minimum cost for a given read write pattern, has been shown to be NPcomplete (see [Wolfson and Milo 1991]) for general graph toplogies even in the centralized case. Thus we first define and analyze the ADR algorithm for tree networks. The analysis is theoretical and experimental. Theoretically, we show that the ADR algorithm is convergent optimal in the following sense. Assume that the read write ....

....topology) is a management hierarchy, in which some of the processors in the communication network are designated as managers, and they form a hierarchy. Net mate, a network management system that we are currently developing (see [Dupuy et al. 1991a; Dupuy et al. 1991b; Sengupta et al. 1990; Wolfson et al. 1991]) employs such a management hierarchy. Net mate s purpose is to provide software tools for the detection of faults and recovery from them in very large communication networks. In Net mate, an object may be transferred from a manager to its superior, or from the superior to the manager, and its ....

[Article contains additional citation context not shown here]

Wolfson, O. and Milo, A. 1991. The multicast policy and its relationship to replicated data placement. ACM Trans. Database Syst. 16, 1.

....Additionally, the model there does not allow concurrent requests, and it requires centralized decision making by a processor that is aware of all the requests in the network. In contrast, our algorithms are distributed, and allow concurrent read write requests. Static allocation was studied in [34, 13]. These works address the following file allocation problem. They assume that the read write pattern at each processor is known a priori or it can be estimated, and they find the optimal static allocation scheme. However, works on the file allocation problem do not compare static and dynamic ....

O. Wolfson and A. Milo, "The Multicast Policy and Its Relationship to Replicated Data Placement", ACM TODS, 16 (1), 1991.

....replicated data are increased availability and performance. No other work of which we are aware, quantitatively compares static and dynamic allocation, while considering both purposes of data replication and treating I O and communication costs in a unified model. Static allocation was studied in [26, 9], however not from an online algorithmic point of view. In other words, these works address the following file allocation problem. They assume that the read write pattern at each processor is known a priori and they find the optimal static allocation scheme. However, works on file allocation ....

O. Wolfson and A. Milo, "The Multicast Policy and Its Relationship to Replicated Data Placement ", ACM TODS, 16 (1), 1991.

....works on replicated data consider the static problem of the replication, namely establishing a priori a replication scheme that will optimize performance, but will remain fixed at runtime. This is called the file allocation problem, and it has been studied extensively in the literature (see [11] and [12] for a survey) In contrast, our algorithm does not keep the replication scheme constant during runtime. Another approach to improve the performance in a replicated distributed database, which also assumes a static replication scheme, is to relax the serializability requirement. Works on ....

O. Wolfson and A. Milo, The Multicast Policy and Its Relationship to Replicated Data Placement, ACM TODS, 16 (1), 1991.

....works on replicated data consider the static problem of replication, namely establishing a priori a replication scheme that will optimize performance, but will remain fixed at runtime. This is called the file allocation problem, and it has been studied extensively in the literature (see [WM], and [DF] for a survey) In contrast, in this paper we discussed dynamic replication. Another approach to improve the performance in a replicated distributed database is to relax the serializability requirement. Works on quasi copies ( ABG1, ABG2, BG] lazy replication ( LLS] and bounded ....

O. Wolfson and A. Milo, "The Multicast Policy and Its Relationship to Replicated Data Placement", ACM Transactions on Database Systems, 16:1, 1991.

....works on replicated data consider the static problem of replication, namely establishing a priori an allocation scheme that will optimize performance, but will remain fixed at runtime. This is called the file allocation problem, and it has been studied extensively in the literature (see [12, 16] for a survey) These works assume that access patterns are known in advance. We feel that this is not always a realistic assumption, particularly in a mobile computing environment. The works on quasi copies ( 17] also assume that the allocation scheme, particularly the divergence threshold, is ....

O. Wolfson and A. Milo, "The Multicast Policy and Its Relationship to Replicated Data Placement ", ACM TODS, 16 (1), 1991.

....Additionally, the model there does not allow concurrent requests, and it requires centralized decision making by a processor that is aware of all the requests in the network. In contrast, our algorithms are distributed, and allow concurrent read write requests. Static allocation was studied in [37, 14]. These works address the following file allocation problem. They assume that the read write pattern at each processor is known a priori or it can be estimated, and they find the optimal static allocation scheme. However, works on the file allocation problem do not compare static and dynamic ....

O. Wolfson and A. Milo, "The Multicast Policy and Its Relationship to Replicated Data Placement", ACM TODS, 16 (1), 1991.

No context found.

O. Wolfson and A. Milo, "The multicast policy and its relationship to replicated data placement," ACM Trans. Database Syst., vol. 16, no. 1, pp. 181--205, Mar. 1991.

No context found.

O. Wolfson and A. Milo, "The multicast policy and its relationship to replicated data placement," ACM Transactions on Database Systems, vol. 16, no. 1, pp. 181--205, Mar. 1991.

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