W. Hong and M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. In PDIS, 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....have implemented these algorithms in the Cohera federated database system [25] and we present experimental results on a set of modified TPC H benchmark queries. Our experimental results, somewhat surprisingly, suggest that the simple technique of breaking the optimization process into two phases [26] first finding the best query plan for a single machine and then scheduling it across the federation based on run time conditions works very well in the presence of fluctuations in the loads on the underlying data sources and the communication costs, as long as the physical database design ....

....factor in the message cost, the total communication cost may not necessarily decrease. Plan Space : 33] discusses the plan space explored by this algorithm. It will be a subspace of the plan space explored by the exhaustive algorithm. 3.4. Two phase Optimization Two phase optimization [26] has been used extensively [9, 19] in distributed and parallel query optimization mainly because of its simplicity and the ease of implementation. This algorithm works in two phases : The techniques described in [33] based on minimum selectivity, etc. can be applied orthogonally. However, ....

[Article contains additional citation context not shown here]

W. Hong and M. Stonebraker. Optimization of parallel query execution plans in xprs. In PDIS, 1991.

....amount of time spent in query optimization grows linearly with the number of buffer sizes for which the query is optimized, which may be prohibitive. Also, as has been pointed out elsewhere [GW89] this approach may work for one or two parameters, but would not scale up. In a more recent reference [HS91], the assumption is made that the buffer size is greater than the minimum required for efficient execution of hash join. Based on that assumption, experimental evidence is provided that the optimal plan is in general insensitive to buffer size. Hence, an enhanced version of a conventional query ....

....and not only buffer size. In that respect, the work of Graefe and Ward is also general [GW89] Second, we develop complete parametric query optimization algorithms that produce multiple plans as output. These algorithms are not based on any assumptions like those made in the XPRS project [HS91], so they are much more generally applicable. Third, the experimental results of these algorithms on the buffer size parameter show that generality is not achieved at the expense of efficiency or output quality. Hence, we expect that these algorithms can easily be incorporated in the systems ....

Hong W, Stonebraker M (1991) Optimization of parallel query execution plans in xprs. In: Proc. of the 1st International PDIS Conference. Miami, Fla., pp 218--225

....because the data is not tightly controlled or exclusively used by the data integration system. For example, query optimization is difficult in the absence of data source statistics. This subjects has been the focus of adaptive query processing research discussed elsewhere (e.g. RS86,WA91,HS93,UF00,HH99,IFF 99, UF01,UFA98,KD98,AH00] However, adaptive query processing research has generally focused on relational query processing, not on XML. One of the major advantages offered by relational query processing has been a pipelined execution model in which new tuples can be read ....

....satisfies an important desideratum for interactive data integration applications. A single pipeline provides the most opportunities for exploiting parallelism and for flexibly scheduling the processing of tuples. This enables the use of techniques such as the pipelined hash join [RS86, WA91,HS93,UF00,HH99,IFF 99] as well as eddies [AH00] Pipelining and adaptive query processing techniques have largely been confined to the relational data model. One of the contributions of this paper is a new XML query processing architecture that emphasizes pipelining the XML data streaming into ....

W. Hong and M. Stonebraker. Optimization of parallel query execution plans in XPRS. Distributed and Parallel Databases, 1(1):9--32, 1993.

....cluster based parallel computing architecture in which each processing node (or site) has a private CPU, memory, and disk, and is connected to all other nodes via a highbandwidth, low latency network. 2.1. Previous Approach A popular approach for parallel database queries consists of two phases [13]. First, a query optimizer generates a static, sequential query plan based on fixed cost models and previously computed statistics. Second, at execution time, this plan is parallelized based on the current runtime characteristics of the system. In the second phase, the degree of parallelism (or ....

W. Hong and M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. In PDIS, 1991.

....cluster based parallel computing architecture in which each processing node (or site) has a private CPU, memory, and disk, and is connected to all other nodes via a highbandwidth, low latency network. 2.1. Previous Approach A popular approach for parallel database queries consists of two phases [14]. First, a query optimizer generates a static, sequential query plan based on fixed cost models and previously computed statistics. Second, at execution time, this plan is parallelized based on the current runtime characteristics of the system. In the second phase, the degree of parallelism (or ....

W. Hong and M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. In PDIS, 1991.

....interfaces tend to generate complex queries that may contain larger numbers of joins between relations. Consequently, the development of execution strategies for the parallel evaluation of multi join queries has drawn the attention of the scientific community. A number of strategies was proposed [CLY92,CYW92, HoS91,HCY94,ScD90] and their performance was evaluated via simulation. However, no comparative experimental performance evaluation is available. This paper describes the proposed strategies in a common framework. Four strategies are implemented on PRISMA DB and a comparative performance evaluation is done. The ....

....to optimize towards minimal total costs. Rather, the exploitation of parallelism has to be taken into account as well. However, the search space that results if all possible trees and all possible parallelizations for these trees are taken into account is gigantic. To overcome these problems, [HoS91] proposes a two phase optimization strategy for multi join queries. The first phase chooses the tree that has the lowest total execution costs and the second phase finds a suitable parallelization for this tree. Although not all researchers agree on this assumption [SrE93] this paper will adopt ....

W. Hong & M. Stonebraker, "Optimization of parallel query execution plans in XPRS," in Proc lstPDIS Conf, Miami Beach, Florida, USA, December 1991.

....German Research Council under contract DFG Fr 1142 1 1. DB2 6000 PE [11] EDS [6] Gamma [9] and Prisma DB [1] or on shared disk, e.g. Rdb on a VAXcluster [19] and Oracle on nCUBE [18] Only a few systems are based on shared everything, e.g. the first version of DBS3 [2] Volcano [14] and XPRS [16]. On the other hand, new trends in parallel hardware develop towards a different direction. About 30 powerful processors are connected to main memory with a new, very fast bus technology. The system can be tuned as a database server by adding multiple I O processors connected to several disk ....

....determine which kind of parallelism to use for query processing. Database machines research concentrated on intra operator parallelism. Most commercial database systems have focused on inter query parallelism, so far [15, 27] Recently, the use of inter operator parallelism has been investigated [8, 16, 24, 25, 26, 29, 32]. Pipelining parallelism is of particular interest. In [24] Schneider and DeWitt study the effect of pipelining on a right deep tree of hash join operators in detail. The evaluation of queries is split into two phases. First, the inner relations are read from disk, and hash tables are built in ....

W. Hong and M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. Distr. and Parallel Databases, Volume 1, Number 1, 1993.

....This approach has two main advantages: communication is very fast and load balancing is simple. The disadvantage of shared everything is its limited scalability due to contention on the interconnection system. Examples for this architecture are DBS3 [BCL93] Volcano [Gra94] and XPRS [HS93] Shared nothing: Each processor accesses only its dedicated memory and disks. Processors communicate by message passing through an interconnection network. The drawbacks of sharednothing are relatively slow communication and difficult load balancing. The advantage of sharednothing is its ....

....which kind of parallelism to use for query processing. Database machines research concentrated on intra operator parallelism. Most commercial database systems have focused on inter query parallelism, so far. Recently, the use of inter operator parallelism has been investigated [SD90, WA91, CLYY92, HS93, SYT93, SE93, ZZBS93] Pipelining parallelism is of particular interest. In [SD90] Schneider and DeWitt study the effect of pipelining on a right deep tree of hash join operators in detail. The evaluation of queries is split in two phases. First, the inner relations are read from disk, and hash ....

W. Hong and M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. Distr. and Parallel Databases, 1(1), 1993.

....architectures, e.g. Bubba [BAC 90] DB2 6000 [Fec95] EDS [BHL 95] Gamma 90] and Prisma DB [AvdBF 92] or on shared disk, e.g. Rdb on a VAXcluster and Oracle on nCUBE. Only a few systems are based on shared everything, e.g. first version of DBS3 [BCL93] Volcano [Gra94] and XPRS [HS93] On the other side, new trends in parallel hardware develop towards a different direction. Examples for this trend are Sun s new Ultra Enterprise X000 Server or Silicon Graphics Power Challenge product generation: about 30 powerful processors and main memory are connected by a fast bus. The ....

....which kind of parallelism to use for query processing. Database machines research concentrated on intra operator parallelism. Most commercial database systems have focused on inter query parallelism, so far. Recently, the use of inter operator parallelism has been investigated [SD90, WA91, CLYY92, HS93, SYT93, SE93, ZZBS93] Pipelining parallelism is of particular interest. In [SD90] Schneider and DeWitt study the effect of pipelining on a right deep tree of hash join operators in detail. The evaluation of queries is split into two phases. First, the inner relations are read from disk, and hash ....

W. Hong and M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. Distr. and Parallel Databases, 1(1), 1993.

....plans and are used to combine data from multiple sources. Second, join operators lead to complex scheduling problems since they proceed in more than one operational stage. 2 Overview of Pipelined Hash Joins The earliest work in non blocking pipelined hash joins is the Symmetric Hash Join (SHJ) HS93, WA90, WA91] which was designed to allow a high degree of pipelining in parallel databases. SHJ builds two hash tables, one for each source. When a tuple arrives on one of the inputs, it is rst inserted into the hash table for that input, and then immediately used to probe the hash table of the ....

W. Hong, M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. Distributed and Parallel Databases, 1(1):9-32, 1993.

....behavior through input rates, and modeling network tra#c as Poisson random processes, has appeared in many contexts, including [2] although to our knowledge it has not been applied in the context of query optimization. A lot of work has been carried out in the areas of non blocking algorithms [5, 14, 16], which aim at producing plans that do not block their execution because of slow input streams. These algorithms are symmetric in the sense that they do not assign di#erent roles to the participating inputs (i.e. there is no distinction between outer and inner streams) Our framework employs such ....

W. Hong and M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. Distributed and Parallel Databases, 1(1):9--32, 1993.

....scheduling algorithm. In Section 6, we present a comprehensive performance analysis. Section 7 contains a discussion of the design decisions. We conclude the paper with Section 8. 2. Related Work Parallel query processing has been studied in a large variety of facets, see e.g. PMC 90, DG92, HS93, WFA95, Gra95] Most of related work in this eld concentrated on possibilities to speedup highly complex queries with long running times. Approaches as taken in [HM94] and [GI96] suggest a decomposition of the query plans into sub plans which are then executed in parallel on di erent nodes of ....

W. Hong and M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. Distributed and Parallel Databases, 1(1):9-32, 1993.

....which kind of parallelism to use for query processing. Database machines research concentrated on intra operator parallelism. Most commercial database systems have focused on inter query parallelism, so far [Gra95, Val93] Recently, the use of inter operator parallelism has been investigated [CLYY92, HS93, SD90, SYT93, SE93, WA91, ZZBS93]. Pipelining parallelism is of particular interest. In [SD90] Schneider and DeWitt study the effect of pipelining on a right deep tree of hash join operators in detail. The evaluation of queries is split into two distinct phases. First, the inner relations are read from disk, and hash tables are ....

W. Hong and M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. Distr. and Parallel Databases, 1(1), 1993.

....and Phrases: parallel and distributed databases, parallel query processing, dynamic load balancing, efficient resource utilization Note: Funded by the HPCN IMPACT project. 1. Introduction Query optimization in parallel database systems is, following a common approach, split into two phases [11, 7]: sequential optimization and parallelization. The former involves query rewrites and join ordering to arrive at an optimal sequential query evaluation plan (QEP) The latter deals with mapping a sequential QEP to a parallel execution environment. The final result is a parallel query execution ....

W. Hong and M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. In Proc. Int'l. Conf. on Parallel and Distr. Inf. Sys., Miami Beach, FL, USA, December 1991.

....that other strategies are impractical in our environment. Other strategies require an arbitrarily large amount of current global knowledge regarding fragment composition and location. This use of standard optimizer technology as a basis for subsequent refinement is similar to that proposed in [HONG91]. In the discussion that follows, the word site generally indicates the Mariposa query processing engine running on a particular computer. The words query plan and query tree are used interchangably. As with many query processing systems, Mariposa implements query execution plans as trees of ....

Hong, W. and Stonebraker, M., "Optimization of Parallel Query Execution Plans in XPRS," Proc. 1st Int. Conf. on Parallel and Distributed Info. Sys., Miami Beach, FL, Dec. 1991.

No context found.

W. Hong and M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. In PDIS, 1991.

No context found.

W. Hong and M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. Distributed and Parallel Databases, 1(1):9--32, 1993.

No context found.

W. Hong, M. Stonebraker, Optimization of Parallel Query Execution Plans in XPRS, Distributed and Parallel Databases, 1(1), January 1993.

No context found.

W. Hong and M. Stonebraker. Optimization of parallel query execution plans in xprs. In ICPDIS, pages 218--225, 1991.

No context found.

W. Hong and M. Stonebraker. Optimization of parallel query execution plans in xprs. In ICPDIS, pages 218--225, 1991.

No context found.

W. Hong and M. Stonebraker. Optimization of parallel query execution plans in XPRS. Distributed and Parallel Databases, 1(1), Jan. 1993.

No context found.

W. Hong and M. Stonebraker. Optimization of parallel query execution plans in XPRS. Distributed and Parallel Databases, 1(1), Jan. 1993.

No context found.

W. Hong and M. Stonebraker. Optimization of parallel query execution plans in XPRS. Distributed and Parallel Databases, 1(1):9--32, 1993.

No context found.

W. Hong and M. Stonebraker. Optimization of parallel query execution plans in XPRS. Distributed and Parallel Databases, 1(1):9--32, 1993.

No context found.

W. Hong and M. Stonebraker. Optimization of Parallel Query Execution Plans in XPRS. Distributed and Parallel Databases, 1(1):9-32, January 1993.

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