Leonidas Fegaras, David Maier, and Tim Sheard. Specifying rule-based query optimizers in a reflective framework. In Proceedings of 3rd International Conference on Deductive and Object Oriented Databases, Scottsdale, Arizona, December 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....that is as friendly to the optimizer as possible. Otherwise, we would be forced to write unnecessarily complicated codes for the optimizer and it would be messy to implement even the simplest optimization rules. We originally planned to build an optimizer generator based on some ideas from [12, 8]. However, as can be seen later in Section 5, the translation of CPL into the monad based representation makes it possible for us to develop our optimizer, rules, and strategies in so simple a manner that an optimizer generator becomes unnecessary. In fact, optimization rules can some time be ....

....evaluation mode assigment, access priority assignment, join migration, and so on. We originally planned to build an optimizer generator, in the same sense as a compiler generator like Yacc, for our system. However, it is evident from the examples in this report an optimizer generator like [8] or [12] would probably not simplify thing very much. The idea of rule based optimization and extensible optimizer is not new [8, 12, 3, 10, 19, etc. However, it is worth pointing out that this report stands out in several ways. First, it presents a simple technique for generating rule ....

Leonidas Fegaras, David Maier, and Tim Sheard. Specifying rule-based query optimizers in a reflective framework. In Proceedings of 3rd International Conference on Deductive and Object Oriented Databases, Scottsdale, Arizona, December 1993.

....the optimization process with the information necessary for its termination. The cost model description is not part of this paper. Search strategies and search space pruning have usually to be provided by the used optimizer generator or have to be implemented in it by hand . We prefer CRML [FMS93] and therefore have to implement the search strategies by ourselves. Finally, the chosen execution plan is mapped into source code of the object oriented database system ObjectStore [OHMS92] for performance tests. We use ObjectStore purely as a storage subsystem to get the possibility to compare ....

....process. The first prototype implementation of the CROQUE OQL parser, both algebras, the components building the interface to ObjectStore (code generation from execution plans) and parts needed by the optimization process is just finished. In the next step, the optimizer generator CRML [FMS93] will be used to implement the search engine (search strategies and search space pruning) The cost model has to be implemented therefore, too. At the same time, we will look for interesting algebraic equivalences (rewriting rules) and more algorithms implementing the physical operations. ....

L. Fegaras, D. Maier, and T. Sheard. Specifying Rule--based Query Optimizers in a Reflective Framework. In Proceedings of the 3rd Int'l Conference on Deductive and Object-Oriented Databases, pages 146--168, New York, December 1993. Springer.

....queries tend to be small (compared with functional programs for example) Still to write about: ffl Naturally Embedded Query Languages [BTBW92] ffl Complexity of Translations [JRSB85] 7. 2 Comparing COKO with Query Rewrites of Rule Based Optimizers Many rule based systems (e.g. Loh88] FMS93] use rules to map algebraic operators to plan level implementations. This was the intended purpose of rules when first proposed ( Fre87] and also characterizes one of the two kinds of rules used in Exodus [CDG 90] and Starburst [HFLP89] KOLA and COKO express query rewrites (i.e. ....

Leonidas Fegaras, David Maier, and Tim Sheard. Specifying rule-based query optimizers in a reflective framework. In Stefano Ceri, Katsumi Tanaka, and Shalom Tsur, editors, Proceedings of the International Conference on Deductive and Object-Oriented Databases, pages 146--168, 1993.

....plan. In effect, our search space only consists of one plan dimension. This shape of search space results in the need to adapt the used common search strategies. 2. 2 Adaptation of Common Search Strategies Our first prototype has been implemented using the optimizer generator described in [FMS93] This optimizer generator is based on CRML [She93, HS93] an extension of the functional language SML [Har93, MTH90, Pau96] In [SMK93] the well known search algorithms are divided into four categories: deterministic, random based, genetic, and hybrid algorithms. We started investigating ....

....to change the generic strategies into algorithms for every kind of search space shape. Our first prototype comprising the random based search strategies random walk, iterative improvement, simulated annealing, and two phase optimization has been implemented using an optimizer generator [FMS93] based on CRML as described in [Ros96, Ill96] 6 The investigations done using our implementations led to the following observations according to the shape of given search spaces. ffl RW is not really satisfying in most cases. ffl Search spaces containing some plateaus of the same cost value ....

L. Fegaras, D. Maier, and T. Sheard. Specifying Rule--based Query Optimizers in a Reflective Framework. In Proc. of the 3rd Int. Conference on Deductive and Object-Oriented Databases, pages 146--168, New York, December 1993. Springer.

....process, the logical and the physical algebra, we discuss a more complex example in Section 6. Search strategies and search space pruning usually have to be provided by an optimizer generator or to be implemented manually. We prefer a generator based on CRML (Compile time Reflective ML [FMS93, HS93, She93] and therefore are able to implement search strategies by ourselves, controlling the search. This allows us to compare different strategies and to find the most appropriate one for different kinds of queries (e.g. dependent on the search space size) this work is in progress, first ....

L. Fegaras, D. Maier, and T. Sheard. Specifying Rule--based Query Optimizers in a Reflective Framework. In Proc. of the 3rd Int'l Conference on Deductive and Object-Oriented Databases, pages 146--168, New York, December 1993. Springer.

....into a physical plan. In effect, our search space only consists of one plan dimension. This shape of search space results in the need to adapt the used search strategies. 2. 2 Adaptation of Common Search Strategies Our first prototype has been implemented using the optimizer generator described in [2]. This optimizer generator is based on CRML [23, 9] an extension of the functional language SML [19] In [24] the well known search algorithms are divided into four categories: deterministic, random based, genetic, and hybrid algorithms. We started investigating random walk, iterative ....

....possible to change the generic strategies into algorithms for every kind of search space shape. Our first prototype comprising the random based search strategies random walk, iterative improvement, simulated annealing, and two phase optimization has been implemented using an optimizer generator [2] based on CRML as described in [21, 10] The investigations done using our implementations led to the following observations according to the shape of given search spaces. RW is not really satisfying in most cases. Search spaces containing some plateaus of the same cost value are not good ....

L. Fegaras, D. Maier, and T. Sheard. Specifying Rule--based Query Optimizers in a Reflective Framework. In Proc. of the 3rd Int. Conference on Deductive and Object-Oriented Databases, pages 146--168, New York, December 1993. Springer.

....control of the search space, it is unconventional and requires careful attention on the part of the DBI to ensure that no valid operator trees are accidently left out of the search space. We will discuss the Volcano optimizer generator in greater detail in Chapter 2. Fegaras, Maier, and Sheard [24] describe a declarative optimizer generator framework for specifying query optimizers. The premise in their work is that much of the specification in current optimizers consists of procedurally defined actions; making these actions declarative results in a cleaner specification language. To this ....

Leonidas Fegaras, David Maier, and Tim Sheard. Specifying rule-based query optimizers in a reflective framework. In Proceedings Third International Conference on Deductive and Object-Oriented Databases, pages 146--168, Phoenix, December 1993.

....and bag languages in [LW94c] see also [GM93] HKM93, HK94] show how to encode related database languages in the simply typed lambda calculus. A central issue is the optimization of collection type queries, see [Sch86, Osb88, Col89, HN91, LD91, LD92, Sar92] and [BK93, ACM93, CV93, CS93, Feg94, FMS93] Some connections between the s.r. monad and comprehension languages and equational transformations leading to optimizations are discussed in [BTBN91, BBW92, Won93] The important extension to object identifiers, which also allows recursive type support is studied in [AK89] 3 Acknowledgements ....

Leonidas Fegaras, David Maier, and Tim Sheard. Specifying rule-based query optimizers in a reflective framework. In Proceedings of 3rd International Conference on Deductive and Object Oriented Databases, Scottsdale, Arizona, December 1993.

....new nodes and not just new trees, and therefore are inexpressible with rule languages based solely on unification. Many rule based systems (e.g. 26] use rules to map algebraic operators to plan level implementations. The transformations addressed do not consider rewriting at the source level. [16] has similar motivations to ours in that they attempt to remove code fragments that appear in rules. Like [26] however, their work primarily addresses rules that express source to physical transformations. They replace head routines with declarative preconditions that test the values of attributes ....

L. Fegaras, D. Maier, and T. Sheard. Specifying rule-based query optimizers in a reflective framework. In S. Ceri, K. Tanaka, and S. Tsur, editors, Proceedings of the International Conference on Deductive and Object-Oriented Databases, pages 146--168, 1993.

....of rewriting. STARS also have required and available properties, which correspond very closely to the notion of inherited and synthesized attributes in attribute grammars. The attribute grammar evaluation mechanism is simplified tremendously because there are only two attributes. Fegaras, et al.[31] have demonstrated a query optimizer generator which uses compile time reflection in ML. Rewriting is controlled via a rule based term rewriting system that has the flavor of attribute grammar evaluation. The attributes from the term rewriting system are used for context control. The rules are ....

Leonidas Fegaras, David Maier, and Tim Sheard. Specifying rule-based query optimizers in a reflective framework. In Proceedings of the The Third International Conference on Deductive and Object-Oriented Databases, Scottsdale, Arizona, December 1993.

....the monoid comprehension calculus. A more formal treatment is presented elsewhere [10, 9] Section 3 describes the optimizer specification language and the optimizer generator that converts optimizer specification into executable code. It is based on our earlier work on query optimizer generators [11]. The real contribution of this paper is given in Section 4. This section presents the logical rules for unnesting queries and for transforming path expressions into joins. M zero unit(x) merge C I list [ x] set fg fxg [ CI bag ffgg ffxgg ] C M zero unit(x) merge C I sum 0 x C max ....

.... maps OPTL specification into executable code (C code) This code is very efficient, because patterns in OPTL are compiled, not interpreted, using standard pattern decomposition techniques [14] The language and the system described in this section extends our earlier work on optimizer generators [11] in two ways: it supports more powerful patterns and it is more user friendly. Terms in OPTL have type Expr. Expr is a tree like data structure which is used for representing both algebraic forms and execution plans. To make the task of writing these term constructions less tedious, OPTL extends ....

[Article contains additional citation context not shown here]

L. Fegaras, D. Maier, and T. Sheard. Specifying Rule-based Query Optimizers in a Reflective Framework. Deductive and Object-Oriented Databases, Phoenix, Arizona, pp 146--168, December 1993. SpringerVerlag, LNCS 461.

....described in this paper. It is described in detail elsewhere [Feg97] Our OQL optimizer is expressed in a very powerful optimizer specification language, called OPTL, and is implemented in a flexible optimization framework, called OPTGEN, which extends our earlier work on optimizer generators [FMS93] OPTL is a language for specifying query optimizers that captures a large portion of the optimizer specification information in a declarative manner. It extends C with a number of term manipulation constructs and with a rule language for specifying query transformations. OPTGEN is a C ....

L. Fegaras, D. Maier, and T. Sheard. Specifying Rule-based Query Optimizers in a Reflective Framework. Deductive and Object-Oriented Databases, Phoenix, Arizona, pages 146--168, December 1993. SpringerVerlag, LNCS 461.

.... in the literature for optimizing large relational queries, even though most of them deal with acyclic query graphs and generate deep left join trees only [8] Even though there are many proposals for optimizers that consider different orders for OODB algebraic operators (see for example [9] and [5]) to our knowledge, none of these methods is actually non exponential. As it is widely believed now in the database community, the goal of an optimizer is not to find the best evaluation plan to execute a query, since this is infeasible for non trivial queries, but to avoid the worst plans. ....

L. Fegaras, D. Maier, and T. Sheard. Specifying Rule-based Query Optimizers in a Reflective Framework. Deductive and Object-Oriented Databases, Phoenix, Arizona, pp 146--168, December 1993. SpringerVerlag, LNCS 461.

....algorithm described in this paper. It is described in detail in [10] Our OQL optimizer is expressed in a very powerful optimizer specification language, called OPTL, and is implemented in a flexible optimization framework, called OPTGEN, which extends our earlier work on optimizer generators [13]. OPTL is a language for specifying query optimizers that captures a large portion of the optimizer specification information in a declarative manner. It extends C with a number of term manipulation constructs and with a rule language for specifying query transformations. OPTGEN is a C ....

L. Fegaras, D. Maier, and T. Sheard. Specifying Rulebased Query Optimizers in a Reflective Framework. Deductive and Object-Oriented Databases, Phoenix, Arizona, pp 146--168, December 1993. Springer-Verlag, LNCS 461.

....interface (syntactic sugar) so that the object language programs look like the programs they represent. It is the right language for expressing programs that manipulate programs. We have used CRML to implement just plain macros , ML language processors, and experimental ML language extensions[2, 3]. An example, described later, is the extension of ML to support attribute grammar computations over ML datatypes. The class of ML extensions that can be rapidly prototyped with CRML is limited only by your imagination. The meta programming paradigm supported by CRML supplies more than just ....

L. Fegaras, D. Maier, and T. Sheard. Specifying rule-based query optimizers in a reflective framework. In Proc. Third Intl. Conf. on Deductive and Object-Oriented Databases, December 1993.

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