Citation Context

...se a promise function and a search strategy with stronger theoretical properties. Since the optimizer generator is not aware of the target data model, we were unable to use search algorithms like A * =-=[HART68]-=- which would have guaranteed the optimal access plan for all queries. Even for the special case of the relational model, we were not able to find a way to calculate the promise of a transformation suc...

Citation Context

... to find the optimal access plan for a query is to simply generate all possible access plans, estimate their respective processing costs, and output the least expensive one. In the System R optimizer =-=[SELI79]-=- this basic strategy is augmented with a pruning technique that deletes all but the cheapest of a set of equivalent subplans at each step of the optimization process. Without pruning, the optimizer wo...

Citation Context

... Introduction In recent years, a number of new data models have been proposed including Daplex [SHIP81], ABE [KLUG82], GEM [ZANI83], GEMSTONE [COPE84], IRIS [LYNG86], Probe [DAYA85, MANO86], Postgres =-=[STON86]-=-, and LDL [TSUR86]. Unfortunately, implementing a database system for a new data model is a difficult and laborious task. The goal of the EXODUS project is to ease the burden of the database implement...

Citation Context

...s almost as good as those produced by exhaustive search, with the search time cut to a small fraction.1. Introduction In recent years, a number of new data models have been proposed including Daplex =-=[SHIP81]-=-, ABE [KLUG82], GEM [ZANI83], GEMSTONE [COPE84], IRIS [LYNG86], Probe [DAYA85, MANO86], Postgres [STON86], and LDL [TSUR86]. Unfortunately, implementing a database system for a new data model is a dif...

Citation Context

...eorem proving, and learning has been used to improve a programs performance, eg. in game playing programs [BARR81]. Most of the query optimization research done to date, as surveyed by Jarke and Koch =-=[JARK84]-=-, deals with relational systems and their extensions. For the designers of previous query optimization programs, the data model has been a given fact. For example, when reordering join trees, [SELI79]...

Citation Context

...ized to be the set of transformations that can be applied to the initial query tree. ����������������������������� 2 OPEN is a standard name for the set of possible next moves in AI search algorithms =-=[BARR81]-=-.6 The general optimization algorithm can now be described as follows. while (OPEN is not empty) Select a transformation from OPEN Apply it to the correct node(s) in MESH Do method selection and cost...

Citation Context

...e search, with the search time cut to a small fraction.1. Introduction In recent years, a number of new data models have been proposed including Daplex [SHIP81], ABE [KLUG82], GEM [ZANI83], GEMSTONE =-=[COPE84]-=-, IRIS [LYNG86], Probe [DAYA85, MANO86], Postgres [STON86], and LDL [TSUR86]. Unfortunately, implementing a database system for a new data model is a difficult and laborious task. The goal of the EXOD...

Citation Context

...future, we plan on investigating generators for user interfaces. An overview of the architecture of EXODUS can be found in [CARE86b]. The design of the storage manager and file system is presented in =-=[CARE86a]-=-. The E programming language is described in [RICH86]. In this paper, we describe the optimizer generator. Until very recently, query optimizers [SELI79, WONG76, KOOI80] have been designed and impleme...

Citation Context

...the optimizer, the access plan, can either be interpreted by a recursive procedure or it can be further transformed. Both approaches have been used successfully in existing database systems. In Gamma =-=[DEWI86]-=-, for example, the operators in the access plan are interpreted (though the predicates themselves are compiled into machine language). In System R [SELI79], the access plan was compiled into machine l...

Citation Context

...ls with relational systems and their extensions. For the designers of previous query optimization programs, the data model has been a given fact. For example, when reordering join trees, [SELI79] and =-=[KOOI80]-=- assume that the order in which joins are executed makes no semantic difference. In the EXODUS optimizer generator, on the other hand, the operators and their semantics are left open, thus allowing th...

Citation Context

...nterfaces. An overview of the architecture of EXODUS can be found in [CARE86b]. The design of the storage manager and file system is presented in [CARE86a]. The E programming language is described in =-=[RICH86]-=-. In this paper, we describe the optimizer generator. Until very recently, query optimizers [SELI79, WONG76, KOOI80] have been designed and implemented with a specific data model and database system i...

Citation Context

...oduced by exhaustive search, with the search time cut to a small fraction.1. Introduction In recent years, a number of new data models have been proposed including Daplex [SHIP81], ABE [KLUG82], GEM =-=[ZANI83]-=-, GEMSTONE [COPE84], IRIS [LYNG86], Probe [DAYA85, MANO86], Postgres [STON86], and LDL [TSUR86]. Unfortunately, implementing a database system for a new data model is a difficult and laborious task. T...

Citation Context

...ecent years, a number of new data models have been proposed including Daplex [SHIP81], ABE [KLUG82], GEM [ZANI83], GEMSTONE [COPE84], IRIS [LYNG86], Probe [DAYA85, MANO86], Postgres [STON86], and LDL =-=[TSUR86]-=-. Unfortunately, implementing a database system for a new data model is a difficult and laborious task. The goal of the EXODUS project is to ease the burden of the database implementor (DBI). EXODUS i...

Citation Context

...n a variety of earlier efforts in the query optimization area. Pioneering work was done in the System R project [ASTR76, SELI79], in the Ingres project [STON76, WONG76, YOUS79] and by Smith and Chang =-=[SMIT75]-=-. Optimization using algebraic identities was first used in compilers for programming languages, but seems to have only been used once for database optimization, in the MICROBE relational distributed ...

Citation Context

...t implements it. file fig1_1.grn Figure 1 As proposed in [CARE85], we initially intended to implement a rule-based optimizer using an AI language like Prolog [WARR77, CLOC81], OPS5 [FORG81], or LOOPS =-=[BOBR83]-=- as those languages provide pattern ����������������������������� 1 A word about the examples in this paper. First, examples based on the relational data model were chosen because they are easily unde...

Citation Context

...the search time cut to a small fraction.1. Introduction In recent years, a number of new data models have been proposed including Daplex [SHIP81], ABE [KLUG82], GEM [ZANI83], GEMSTONE [COPE84], IRIS =-=[LYNG86]-=-, Probe [DAYA85, MANO86], Postgres [STON86], and LDL [TSUR86]. Unfortunately, implementing a database system for a new data model is a difficult and laborious task. The goal of the EXODUS project is t...

Citation Context

...ith the result of the first phase instead of the initial query tree. Finally, the third phase would do work analogous to peep hole optimization in compiler technology, eg. predicate clause reordering =-=[HANA77]-=-. Other assignments of tasks to phases could be designed as well. The idea of phases is quite similar to (actually a generalization of) the idea of a "pilot pass" [ROSE86]. The first real test for the...

Citation Context

...od as those produced by exhaustive search, with the search time cut to a small fraction.1. Introduction In recent years, a number of new data models have been proposed including Daplex [SHIP81], ABE =-=[KLUG82]-=-, GEM [ZANI83], GEMSTONE [COPE84], IRIS [LYNG86], Probe [DAYA85, MANO86], Postgres [STON86], and LDL [TSUR86]. Unfortunately, implementing a database system for a new data model is a difficult and lab...

Citation Context

...to the optimization process. Since EXODUS does not support a single conceptual data model, it would impossible to provide a single optimizer for all target applications. As a solution we hypothesized =-=[CARE85]-=- that if the query optimizer were organized as rule-based system, then as new operators, access methods, etc., were added to the database system, the optimizer could be informed of their properties by...

Citation Context

...pes and procedures defined in the system, and an optimizer generator. In the future, we plan on investigating generators for user interfaces. An overview of the architecture of EXODUS can be found in =-=[CARE86b]-=-. The design of the storage manager and file system is presented in [CARE86a]. The E programming language is described in [RICH86]. In this paper, we describe the optimizer generator. Until very recen...

Citation Context

...sing algebraic identities was first used in compilers for programming languages, but seems to have only been used once for database optimization, in the MICROBE relational distributed database system =-=[NGUY82]-=-. Freytag assumes in his work on21 code generation [FREY85, FREY86] for access plans that query plans for set-oriented data models can be expressed as trees. Search strategies have been used in the a...

Citation Context

... predicate clause reordering [HANA77]. Other assignments of tasks to phases could be designed as well. The idea of phases is quite similar to (actually a generalization of) the idea of a "pilot pass" =-=[ROSE86]-=-. The first real test for the optimizer generator will come when it is used for a real system. The EXODUS project team intends to implement a relational database system. The first real system will be ...

Citation Context

...ced by a method that implements it. file fig1_1.grn Figure 1 As proposed in [CARE85], we initially intended to implement a rule-based optimizer using an AI language like Prolog [WARR77, CLOC81], OPS5 =-=[FORG81]-=-, or LOOPS [BOBR83] as those languages provide pattern ����������������������������� 1 A word about the examples in this paper. First, examples based on the relational data model were chosen because t...

Citation Context

...filescan A filescan BsSMllA SCXIIBsY:*@sFigure 1sAs proposed m [CARE85], we mlhally intended tosimplement a rule-based optmuzer usmg an AI language hkesProlog [wARR77, CLOC81],OPS5 [FORG81], or LOOPSs=-=[BOBR83]-=- as those languages prowde pattern matchmg andsa search engme, and since umficatron can be used elegantlysto bmld new query trees from old ones In addmon. theseslanguages allow augmentation of the rul...

Citation Context

...EY85, FREY86al for access plans that query plans forsset-onented data models can be expressed as treessRecently, Freytag has begun work on deslgnmg arule-basedsophmmmon scheme for the relauonal model =-=[FREY86b]-=-sSearch strateges have been used m the areas of deductionsand theorem provmg, and learnmg has been used to nnorovesa programs perfoniance, eg ii game playmg pm&amss[BARR811sM&t of the query opnrmzahon...

Citation Context

...startmg with the result of the first phasesmstead of the mlhal query tree Fmally, the thnd phaseswould do work analogous to peep hole optmuzation m com-spder technology, eg predtcate clause reordenng =-=[HANA77]-=-sOther assignments of tasks to phases could be designed asswell The idea of phases is qmte slrmlar to (actually a gen-seralization of) ua idea of a “pfiot pass” [ROSE861sThe first real test for the op...

Citation Context

...ionsand theorem provmg, and learnmg has been used to nnorovesa programs perfoniance, eg ii game playmg pm&amss[BARR811sM&t of the query opnrmzahon research done to date,sas surveyed by Jarke and Koch =-=[JARK84]-=-, deals with rela-shonal systems and theu extensions For the designers of pre-sVIOUS query opmzafion programs, the data model has beensa @ven fact For example, when reordermg jam trees,sISELI and [KOO...

Citation Context

...lthsthe search tune cut to a small fractions1 IntroductionsIn recent years, a number of new data models havesbeen proposed mcludmg Daplex [SHIPSl], ABE [KLUG82],sGEM [ZANISJ], GEMSTONE [COPE84], IRIS =-=[LYNG86]-=-,sProbe [DAYA85, MANO863, Postgres [STON86], and LDLs[TSUR861 Unfortunately, Implementmg a database systemsfor a new data model 1s a difficult and labonous task Thesgoal of the EXODUS project 1s to ea...

Citation Context

...BIsto design and expenment ~rlth new data modelssAlgebmc transformanon laws have also been used msthe design and implementation of the opturnzer for the dlsm-sbuted relational database system MICROBE =-=[NGUY82]-=-sThe goal of the MICROBE rule based opmzauon step wassto numm~ze the number of operators and the amount of datasto be shpped between operators A set of transformationsrules was formulated and proven t...

Citation Context

...s1 IntroductionsIn recent years, a number of new data models havesbeen proposed mcludmg Daplex [SHIPSl], ABE [KLUG82],sGEM [ZANISJ], GEMSTONE [COPE84], IRIS [LYNG86],sProbe [DAYA85, MANO863, Postgres =-=[STON86]-=-, and LDLs[TSUR861 Unfortunately, Implementmg a database systemsfor a new data model 1s a difficult and labonous task Thesgoal of the EXODUS project 1s to ease the burden of thesdatabase unplementor (...

Citation Context

...ntsshortcommgs are The second real test Hrlll be when we setsout to design an opmzer for one of the recently proposedsnew data models, eg ABE [KLUG82], Daplex [SHIPSl],sProbe [DAYA85, MANO861, or LDL =-=[TSUR86]-=-sFmally, we realize that the optumzer generator worksslargely on the syntacac level of the algebra The semanttcssof the data model are left to the DBI’s code Thts has thesadvantage of allowmg the DBI ...