C.S. Ellis. Distributed data structures: A case study. IEEE Transactions on Computing, C- 34(12):1178-1185, 1985.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....linked data structures. Multi version Memory permits only a single update to occur on a replicated node at any point in time (analogous to value logging [23, 24] in transaction systems) Our algorithm permits concurrent updates on replicated nodes (analogous to transition logging [23, 24] Ellis [25] has proposed algorithms for a distributed hash table. The directories of the table are replicated among several sites, and the data buckets are distributed among several sites. The hash table is a shallow search structure, so every update to the index structure must be distributed to every copy ....

....many of the modifying suboperations commute. Several authors have studied the issue of concurrency control on abstract data types when some operations may commute [31, 32, 33, 34, 35] but in the context of a transaction processing system. Our algorithms are similar to those described by Ellis [25] to maintain the replicated directories of the distributed hash tables in so that it is not always necessary that all 2.2 Integrating Concurrency Control with Replica Coherency suboperations are performed in the same order at all nodes. For example, insert suboperations may be performed out of ....

[Article contains additional citation context not shown here]

C.S. Ellis. Distributed data structures: A case study. IEEE Transactions on Computing, C- 34(12):1178-1185, 1985.

....linear hashing method that s particularly useful for main memory databases is discussed in [17] The paper addresses the problem of maintaining local copies of the centralized variables, and discusses various recovery mechanisms. Distributed memory data structures have been proposed by Ellis [3], Severance [17] Peleg [14] Colbrook et al. 2] and Johnson and Colbrook [6] Ellis [3] has proposed a distributed extendible hashing technique, that uses techniques similar to the ones we use here. Colbrook et al. 2] have proposed a pipelined distributed B tree, where each level of the tree is ....

....in [17] The paper addresses the problem of maintaining local copies of the centralized variables, and discusses various recovery mechanisms. Distributed memory data structures have been proposed by Ellis [3] Severance [17] Peleg [14] Colbrook et al. 2] and Johnson and Colbrook [6] Ellis [3] has proposed a distributed extendible hashing technique, that uses techniques similar to the ones we use here. Colbrook et al. 2] have proposed a pipelined distributed B tree, where each level of the tree is maintained by a different processor. The parallelism achieved is limited by the height ....

Ellis S. C. Distributed Data Structures: A Case Study, IEEE Transactions on Computers, Vol. C-34, No. 12, December 1985. 16

....linked data structures. Multi version Memory permits only a single update to occur on a replicated node at any point in time (analogous to value logging [23, 24] in transaction systems) Our algorithm permits concurrent updates on replicated nodes (analogous to transition logging [23, 24] Ellis [25] has proposed algorithms for a distributed hash table. The directories of the table are replicated among several sites, and the data buckets are distributed among several sites. The hash table 4 2 THE DB TREE, A CONCURRENT DISTRIBUTED B TREE is a shallow search structure, so every update to the ....

....many of the modifying suboperations commute. Several authors have studied the issue of concurrency control on abstract data types when some operations may commute [31, 32, 33, 34, 35] but in the context of a transaction processing system. Our algorithms are similar to those described by Ellis [25] to maintain the replicated directories of the distributed hash tables in so that it is not always necessary that all 2.2 Integrating Concurrency Control with Replica Coherency 11 suboperations are performed in the same order at all nodes. For example, insert suboperations may be performed out of ....

[Article contains additional citation context not shown here]

C.S. Ellis. Distributed data structures: A case study. IEEE Transactions on Computing, C34 (12):1178--1185, 1985.

....eventually generate data that need to be stored in the global structure and may be needed by other processors. In this paper, we deal specifically with priority queue data structures. Compared with data structures used for other applications, such as Distributed Databases or Dictionary machines [1, 3, 9, 11], the main difference arises from the natural centralized control required to detect the site which holds the item with the largest priority. For this reason, the tenet that a tree topology is not appropriate (by creating hot spots and bottlenecks) no longer makes sense for distributed priority ....

C.S. Ellis. Distributed data structures: A case study. IEEE Transactions on Computers, C-34(12):1178-- 1185, December 1985.

....memory loads over the sites of the network. We do not consider here implementation with some specific processors managing the data structure for other distributed working processors (our purpose is neither a dictionary machine nor a VLSI implementation; these fields have been intensively explored [1, 7, 20, 23]) In this paper, we deal specifically with priority queue data structures they supports exclusively the operations insert and deletemin. Compared with data structures used for other applications, such as Distributed Databases or Dictionary machines, the main difference arises from the natural ....

C.S. Ellis. Distributed data structures: A case study. IEEE Transactions on Computers, C-34(12):1178--1185, December 1985.

....network. Some related work has been done to implement hash tables. Yen and Bastani [43] have developed algorithms for implementing a hash table on a SIMD parallel computer, such as a CM2. The authors examine the use of chaining, linear probing, and double hashing to handle bucket overflows. Ellis [11] has proposed algorithms for a distributed hash table. The directories of the table are replicated among several sites, and the data buckets are distributed among several sites. Ellis algorithm for maintaining the replicated directories is similar in many ways to our lazy update algorithms. ....

....include [37, 28, 33] The contribution of this work is to present a method for constructing algorithms for distributed search structures. Unlike much of the previous work, our algorithms are explicitly designed for asynchronous distributed systems (or asynchronous parallel processors) While Ellis [11] and Litwin, Neimat, and Schneider [26, 38] have also proposed distributed search structure algorithms with a similar flavor, we present a structure for understanding, designing, and proving correct more complex distributed search structure algorithms. 1.1.2 The dB tree We use the dB tree as a ....

[Article contains additional citation context not shown here]

C.S. Ellis. Distributed data structures: A case study. IEEE Transactions on Computing, C-34(12):1178-- 1185, 1985.

.... Practical and efficient implementations of nonblocking queues are described in [Val95, MS96] There are many constructions of concurrent B trees, mainly for use in databases; see for example [BS77, LY81, Sag85] AVL trees, 2 3 trees, and a distributed extendible hash file have been implemented in [Ell80a, Ell80b, Ell85]. A distributed dictionary structure is studied in [Pel90] 2 Implementing test and set bits In this section we provide a fast implementation of a restricted type of test and set bits from atomic registers using timing assumptions. This implementation will play a central role in the ....

C. S. Ellis. Distributed data structures: A case study. IEEE Transactions on Computers, c-34(12):1178--1185, 1985.

....the hash table directory doesn t change, so only the table entries need to be locked, in the obvious way. Several extensible hashing schemes have been proposed, such as Extendible Hashing [26] and Linear Hashing [54] Extensible hashing algorithms increase the versatility of hash tables. Ellis [21, 22] discusses a concurrent Extendible hashing algorithm that is suitable for distributed systems. In an Extendible hashing table, the hash table directory contains entries keyed only to the first k bits of the hash function. Initially, the directory contains only two entries, keyed to the first bit ....

C.S. Ellis. Distributed data structures: A case study. IEEE Transactions on Computing, C-34(12):1178--1185, 1985.

.... on leaf nodes (we have previously found that never merging nodes results in little loss in space utilization [11] and data balancing on the leaf level is low overhead and effective [14] The methods that we present can be applied to other distributed search structures, such as hash tables [5]. 1.1 The dB tree We initiated our study of distributed search structures by examining the design of a highly concurrent distributed B tree. A B tree is a multi ary tree in which every path from the root to the leaf is the same length. The tree is kept in balance by adjusting the number of ....

.... free at empty policy provides good space utilization [11] and that leaf level data balancing is effective and low overhead [14] We provide a correctness theory for lazy updates, so lazy update techniques can be used to implement lazy updates on other distributed and replicated search structures [5]. Lazy updates, like lazy replication, permit the efficient maintenance of the replicated index nodes. Since little synchronization is required, lazy updates permit concurrent search and modification of a node, and even concurrent modification of a node. Finally, distributed search structures ....

C.S. Ellis. Distributed data structures: A case study. IEEE Transactions on Computing, C-34(12):1178-- 1185, 1985.

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