Edward Sciore and John Seig Jr. A Modular query Optimizer Generator. Proceedings of IEEE Conference on Data Engineering, Los Angeles, California, February 1990.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....to divide a query optimization system into regions, which are responsible for di erent parts of optimization. In this system, the query is passed through these various regions to be optimized. Some examples of these systems include: Mitchell s region based optimizer [9] Sciore s modular optimizer [4], and Kemper s blackboard architecture [1] The region based optimizer creates a hierarchy of regions. The parent region dynamically controls the sequence of regions that the query is passed through. Instead, in the blackboard approach, knowledge sources are responsible for controlling the path in ....

Edward Sciore and John Seig Jr. A Modular query Optimizer Generator. Proceedings of IEEE Conference on Data Engineering, Los Angeles, California, February 1990.

....than making this choice a priori as in the Starburst design. Sciore and Siegcriticized earlier rule based query optimizers and concluded that modularity is a major requirement for extensible query optimization systems, e.g. in the rule set and in the 15 control structures for rule application [ScS90]. The different tasks of query optimization, such as rule application and selectivity estimation, should be encapsulated in separate and cooperating experts. Mitchell et al. recently proposed a very similar approach for query optimization in object oriented database systems [MZD92] While ....

E. Sciore and J. Sieg, A Modular Query Optimizer Generator, Proc. IEEE Int'l. Conf.onData Eng., Los Angeles, CA, February 1990, 146.

....instead) Hence, it may become a basic component of a future VirtualMedia Server . 4. 3 Optimization Rule based transformation and optimization of operator graphs have been studied for more than a decade in the context of extendible database query optimizers [RH87, CZ96] optimizer generators [SS90, GM93], and query optimizers for object oriented databases [VD91] Structurally, filter graphs are nearly identical to operator graphs. Semantically, of course, they are different there exist, however, several analogies: Logic operators (e.g. join) correspond to virtual filters, while the ....

Sciore, E., Sieg, Jr, J.: A Modular Query Optimizer Generator. In: Proc. 6 th Int. Conf. on Data Engineering, 1990, pp. 146--153.

....In this section, the work proposed for this thesis is compared to related work in the field. The literature review is divided by category. 7.1 Comparing KOLA with Query Algebras 7.1. 1 Object Query Algebras Still to write about: ffl AQUA [LMS 93] ffl EXTRA [VD91] ffl OFL [GMP95] ffl OGL [SSJ90] ffl GOM [KMP87] ffl Fegaras and Maier [FM95] ffl CROQUE [GKG 97] ffl Steenhagen [SABd94] 7.1.2 Combinator Based Query Algebras We are not the first to suggest that a combinator style for queries facilitates optimizer formalization. EL91] consider the benefits of an FP style ( Bac78] ....

....using its meta rule language as discussed below. Still to write about: ffl Cascades [Gra95, ONK 95] in more detail) ffl CROQUE [HK96] ffl Steenhagen [SABd94] Expressing Complex Query Rewrites with Rule Groups or Modules Many systems (e.g. PHH92] BG92] FG94] MDZ93] LV91] SSJ90] provide some form of meta control language for rules that includes rule grouping and sometimes sequencing. Rule groups can be associated with search strategies that indicate how the rules in a group should be fired. Search strategies are generic (i.e. they do not refer to the rules in any one ....

[Article contains additional citation context not shown here]

Edward Sciore and John Sieg Jr. A modular query optimizer generator. In Proceedings of the 6th International Conference on Data Engineering, pages 146--153, Los Angeles, USA, 1990.

....strategies like materialization or cost based evaluation of semantically equivalent filter graphs. Rule based transformation and optimization of operator graphs have been studied for more than a decade in the context of extendible database query optimizers [RH87, CZ96] optimizer generators [SS90, GM93], and query optimizers for object oriented databases [VD91] Structurally, filter graphs are nearly identical to operator graphs. Semantically, of course, they are different there exist, however, several analogies: Logic operators (e.g. join) correspond to virtual filters, while the ....

Sciore, E., Sieg, Jr, J.: A Modular Query Optimizer Generator. In: Proc. 6 th Int. Conf. on Data Engineering, 1990, pp. 146--153.

....rule specification scheme that is highly error prone and hard to modify in a streamlined manner. Without a mechanism to add or remove sets of rules and automatically generate new rule sets quickly, such systems are extremely limited in the scope and speed of customization. Sciore and Sieg [18] describe an optimizer generator model that allows a DBI to construct a rulebased optimizer using modules. Each module consists of term rewrite rules (with conditions) to transform terms in a relational algebra. Each module has exported and imported interfaces which consist of terms. Each module ....

Edward Sciore and John Sieg, Jr. A modular query optimizer generator. In Proceedings 6th International Conference on Data Engineering, pages 146--153, Los Angeles, February 1990.

....optimizer cannot be implemented using the Cascades Framework. Various architectures have been proposed to allow extensible control over the search strategy of an optimizer. The region based optimizer architecture of Mitchell et al. MDZ93] the modular optimizer architecture by Sciore and Sieg [SJ90] the blackboard architecture of Kemper, Moerkotte and Peithner [KMP93] are all based on the concept of dividing an optimizer into regions that carry out different parts of the optimization. A query then has to pass through these various regions to be optimized. These architectures differ in the ....

....[KMP93] are all based on the concept of dividing an optimizer into regions that carry out different parts of the optimization. A query then has to pass through these various regions to be optimized. These architectures differ in the methods used to pass control between the various regions. In [SJ90] control passes from one region to another in a fixed sequence. MDZ93] uses a hierarchy of regions in which the parent region dynamically controls the sequence of regions through which the query passes while being optimized. In the blackboard approach [KMP93] knowledge sources are responsible ....

Edward Sciore and John Seig Jr. "A Modular Query Optimizer Generator". In Proc. IEEE Conf. on Data Engineering, Los Angeles, California, February 1990.

....the problems introduced by the complexities of the model will require multiple strategies for applying transformations. Current extensible optimizers have concentrated on accommodating extensibility of the data model, new algebraic operators, new transformation rules, and new data access strategies[8, 9, 13, 26]. They are generally based on rewrite rules for a set of operators defined on the bulk types. These rules are applied to a query expression to generate equivalent, but hopefully more efficient, forms of the expression. The transformed expressions can then be evaluated according to a cost model to ....

....these are typically written into the code of the optimizer. They are, therefore, impossible to extend or change without a major rewrite of the system. Optimizer generators can partially address this problem by allowing a control strategy to be defined for an optimizer at the time it is generated[26]. Lanzelotte and Valduriez extend this idea by defining a number of search control strategies in the optimizer and choosing a strategy depending upon the query expression being optimized[17] The architecture we present in this paper allows the definition of any number of control strategies in the ....

[Article contains additional citation context not shown here]

E. Sciore and J. Sieg, Jr., "A Modular Query Optimizer Generator," in Proceedings of the 6th International Conference on Data Engineering, pp. 146--153, 1990.

....differ in all systems [45, 54, 115, 137] Some systems also recognize a need to support new strategies for optimization; i.e. extensibility of the optimization process itself. This is the motivation behind Epoq, presented in Chapter 5. Other approaches to providing this same kind of extensibility [89, 129] are discussed in Section 5.1 where they can be compared with our approach. 2.4 Object Oriented Query Optimization Optimization techniques and results in relational and extensible databases form the basis for current research in object oriented query optimization. In this section we first look ....

....new strategies for manipulating query expressions. The collection of strategies in an optimizer can be extended in response to other extensions in the query system. This strategy extensibility is provided by an Epoq optimizer, and also motivates the approaches taken by the optimizers of [89] and [129]. The Epoq approach to strategy extensibility is provided with a modular architecture in which each module specifies a strategy for query optimization. An Epoq optimizer is a collection of modules, called regions, each of which can manipulate a query expression to discover some subset equivalent ....

[Article contains additional citation context not shown here]

Edward Sciore and John Sieg, Jr. A Modular Query Optimizer Generator. In Proceedings of the 6th International Conference on Data Engineering, pages 146--153, 1990.

....and to control the manner in which a query representation is traversed. Such control is not provided with Gral rules and instead must be expressed using the Gral meta rule language as discussed below. Expressing Complex Query Rewrites with Rule Groups: Many systems (e.g. 15] 1] 7] 13] [17]) provide some form of metacontrol language for rules that includes rule grouping and sometimes sequencing. Rule groups can be associated with search strategies that indicate how the rules in a group should be fired. Search strategies are generic (i.e. they do not refer to the rules in any one ....

....all possible alternatives before comparing them. Some systems permit some variations on exhaustive search (such as prioritizing rules [15] or limiting the number of passes over the query in advance [7] Other systems provide pruning strategies so that only some alternatives are generated (e. g, [17] provide such strategies as branch and boundand simulated annealing) Some systems try to avoid generating poor alternatives by ranking rules (e.g. 15] 13] and [2] or algebraic operators (e.g. 1] to choose a rule to apply next. COKO groups KOLA rules, not to generate alternatives, but to ....

E. Sciore and J. Sieg Jr. A modular query optimizer generator. In Proceedings of the 6th International Conference on Data Engineering, pages 146--153, Los Angeles, USA, 1990.

....Moreover, as we have seen in Chapter 4, it is important to have a compiler that can generate efficient, compact optimizers from a specification constructed using components. None of the proposals discussed in this section describe how that is done, or even if it is possible. Sciore and Sieg [44] describe an optimizer generator model that allows a DBI to construct a rule based optimizer using modules. Each module consists of term rewrite rules, where rewrite rules transform terms in a relational algebra. These rules can have conditions associated with them. Each module has exported and ....

....encapsulation of rule sets in layers, we can foresee each layer encapsulating a rule engine. Thus, the need for a global search strategy disappears, and each layer implements its own search strategy (in addition to its own search space and cost model) This is similar to the frameworks proposed in [39, 40, 44]. One interesting problem is the compaction of layers namely, how can we generate a monolithic rule set with a search strategy that is semantically equivalent to the composition of the search strategies in the different layers Since each layer can have a different rule engine, any compaction ....

Edward Sciore and John Sieg, Jr. A modular query optimizer generator. In Proceedings 6th International Conference on Data Engineering, pages 146--153, Los Angeles, February 1990.

....queries. The kinds of rules used to describe query transformations, and the control over execution of those rules, differ in all systems [5, 7, 16, 19] Some systems also recognize a need to support new strategies for optimization; i.e. extensibility of the optimization process itself [11, 14, 18]. The control we present here is designed to support this kind of extensibility. The Epoq approach to extensible query optimization allows extension of the collection of control strategies that can be used when optimizing a query [14] Each strategy can search some portion of the space of queries ....

....the lowest cost query it can find. Region LC can choose to use this region to attain its own goal, or can choose to send the query through a sequence of modules (CNF and DP, here) each of which has its own strategy for applying its smaller collection of rules to the query (similarly to [5] or [18]) LC could even try both strategies, choosing the best result, or could use the strategies in a pilot pass type of approach. The leaf regions in this example are complicated strategies for the application of transformations to queries. These strategies are controlled by higher level regions. It ....

[Article contains additional citation context not shown here]

Edward Sciore and John Sieg, Jr. A Modular Query Optimizer Generator. In Proceedings of the 6th International Conference on Data Engineering, pages 146--153, 1990.

....Despite the wide spread use of the rule based approach, scant mention can be found discussing design issues for rule languages. Rather, rule languages are usually assumed to be by products of algebra definitions and not considered in and of themselves. An exception is the work of Sciore and Sieg [35], who suggest ways to augment rule languages over variablebased algebras to ensure formulationof a widevariety of rules. Proposed extensions include rule preconditions (expressed in code) and multivariables (abstractions of variable lists) that allow optimization rules to be independent of ....

....the freeness of variables that can no longer be referenced. In short, multivariables and precondition code are intended to address the same problem that we do; variables in query algebras make rules over algebraic representations difficult to express without additional machinery. But whereas [35] add the machinery, we instead remove the variables. As we mentioned earlier, ours is not the first combinatorbased algebra proposed in a database context. 15] and [5] propose an FP style [3] query language. But combinator style languages are difficult for users to master and thus ill suited as ....

E. Sciore and J. S. Jr. A modular query optimizer generator. In Proceedings of the 6th International Conference on Data Engineering, pages 146--153, Los Angeles, USA, 1990.

....database agent into multiple role oriented processes would result in a high overhead for forwarding queries, subQEPs or intermediate results between processing nodes or costly agent context switches. Nevertheless we still use a modular or component based approach for the architecture similar to [19, 21]. Some systems are built as wrappers around existing programs providing access to the local database. They implement a common interface for a group of loosely coupled programs and in addition must solve data model and semantic heterogeneities [23] Such a federated system was proposed in [18] ....

E. Sciore and J. Sieg. A modular query optimizer generator. In Proceedings of the IEEE International Conference on Data Engineering, Los Angeles, CA, Feb. 1990.

....this flexibility is achieved without sacrificing performance. Various architectures have been proposed to allow extensible control over the search strategy of an optimizer. The region based optimizer architecture of Mitchell et al. MDZ93] the modular optimizer architecture by Sciore and Sieg [SJ90] and the blackboard architecture of Kemper, Moerkotte and Peithner [KMP93] are all based on the concept of dividing an optimizer into regions that carry out different parts of the optimization. A query then has to pass through these various regions to be optimized. They differ in the methods ....

....and Peithner [KMP93] are all based on the concept of dividing an optimizer into regions that carry out different parts of the optimization. A query then has to pass through these various regions to be optimized. They differ in the methods used to pass control between the various regions. In [SJ90] control passes from one region to another in a fixed sequence. In [MDZ93] there is a hierarchy of regions in which the parent region dynamically controls the sequence in which the query passes through the various regions while being optimized. In the blackboard approach [KMP93] knowledge ....

Edward Sciore and John Seig Jr. "A Modular Query Optimizer Generator". In Proc. IEEE Conf. on Data Engineering, Los Angeles, California, February 1990.

No context found.

Edward Sciore and John Sieg, Jr., "A Modular Query Optimizer Generator", Proc. 6th Intl. Conf. on Data Engrg., 146-153, 1990.

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