J. Blakeley, W. McKenna, and G. Graefe. Experiences building the Open OODB query optimizer. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 287--295, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....and selectivity estimation formulas, and performance studies have been reported in [SJS01b] The latter paper did not cover the enhancements to Volcano, which are the foci of this paper. While to our knowledge, nobody has enhanced existing optimizers with support for sets, multisets, and lists, [BKG93] reports on experiences from building the query optimizer for Texas Instruments Open OODB system using Volcano. That paper concludes that when building an optimizer, it is essential to have a full optimization framework that includes not only an algebra and its transformations, but also the ....

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences Building the Open OODB Query Optimizer. In Proceedings of ACM SIGMOD, Washington, DC, pp. 287--296 (1993).

....In case we are assembling two trees of operators having partitions of the same original relation, the IDs of the partitions produced by the first tree would be used to probe the sparse B Tree indexes of the partitions in the second. This operator is similar to the materialize operator proposed in [4] for evaluating pointer joins in object oriented database systems. Consider the simple join query considered in the previous section. First consider how the search space of plans is explored for the NSM model. For simplicity we assume that the database has no indexes to use for this query. The ....

J.Blakeley, W.J.McKenna, G.Graefe. Experiences building the open OODB Query Optimizer. Proceedings of ACM SIGMOD 1993.

....tablecan and indexescan are usually parallelised based on the partitioning of the data being read. Thus, even if an attribute sensitive predicate is specified in any of these scan operations, no exchange operator is required to do data repartitioning. The operator materialise, first proposed in [2], is mainly responsible for materialising path expressions by bringing into scope objects of collections referred to in a path expression. The operator reduce is considered as attribute insensitive, as its role is simply to structure the results of a query. However, an exchange operator may be ....

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences building the Open OODB query optimizer. In Proceedings of the ACM SIGMOD, pages 287-296, Washington, DC, USA, 1993.

....A n ] for an extent belonging to object type E. It returns tuples consisting of the object identifier e and projects on the (possibly set valued) attributes A 1 ; A n . The algebraic counterpart of the dot operator in OQL is the expand operator c [KM90] also called, e.g. materialize [BMG93]. It may be used both to retrieve attributes and to invoke member functions of a referenced object. In this paper, we only need the attribute access variant (The operator ffi denotes tuple concatenation and g is a newly introduced attribute) c g : e:a (E) fe ffi [g : e:a] j e 2 Eg To flatten ....

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences building the Open OODB Query Optimizer. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 287--295, Washington, DC, USA, May 1993.

....a lot of work on algebras in the context of object oriented databases. Our work has been influenced mostly by [FM95a] and [FM95b] Monoid homomorphisms are a strong candidate for an XML calculus that we plan to devise in our future work (see Section 9) The object oriented algebra presented in [BMG93] uses a combination of materialize and unnest to access set valued attributes which is close to what our Follow operator does (Section 5.2) Unnest extracts references from sets and materialize resolves these references to in memory objects. Our approach di#ers in two aspects: 1. It centers on ....

Jose A. Blakeley, William J. McKenna, and Goetz Graefe. Experiences building the open OODB query optimizer. In Peter Buneman and Sushil Jajodia, editors, Proceedings of the 1993 ACM SIGMOD International Conference on Management of Data, Washington, D.C., May 26-28, 1993, pages 287--296. ACM Press, 1993.

....TPC H queries in Section 5. Section 6 concludes the paper. 2. Preliminaries In this section we review the concept of a compact representation of the plan space in form of the MEMO structure. This concept was developed by Graefe and DeWitt in the context of transformation based query optimization [4, 5, 1]. Independent of this development, a similar structure has been developed for bottom up enumeration of join trees in Starburst [8] Our technique is based on the MEMO but could be transferred easily to the Starburst enumerator. We will briefly introduce the essential aspects of the MEMO and refer ....

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences Building the Open OODB Query Optimizer. In Proc. of the ACM SIGMOD Int'l. Conf. on Management of Data, Washington, DC, USA, May 1993.

.... operators operators Get Table scan Index scan Join Nested loop Merge join Materialize Materialize Nest Nest Groupby Unnest Unnest Sort Reduce Reduce Union Union Map Map The physical algebra outlined in table 1 is similar to the one described in [6] However, the Materialize operator [1] has been added to this algebra, providing the execution engine with an operator capable of bringing into scope objects from another extent and performing a join between those objects and the input objects, by following the relationships between the input objects and the objects of the hidden ....

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences building the Open OODB query optimizer. In Proceedings of the ACM SIGMOD, pages 287--296, Washington, DC, USA, 1993.

....a set of reasonable execution plans for processing a given query, and selects the one with the minimum estimated cost. One of the most important factors that affects the cost estimation of an execution plan is the intermediate result cardinality the number of objects in an intermediate result [3, 7, 19, 21]. This is because the intermediate result becomes the operand of the next operation in the execution plan, and the cost of the next operation can be estimated mainly based on the number of objects in its operands [18, 19, 21] Hence, accurate estimation of intermediate result cardinality is a ....

....The selectivity of this nested predicate is the ratio of the number of path instances of the j D j A j D i A C i Fig. 4: Condition C i :A i [D j ] A j = val . path C i :A i [D j ] A j that satisfy A j = val to the total number of the path instances. Most of the conventional literature [2, 3, 9, 16, 17] on object oriented databases have no explicit estimation techniques for path selectivity. Kim et al. 11] use the technique proposed by Selinger et al. 19] for estimating the path selectivity as follows: sel(C i :A i [D j ] A j = val ) sel(D j :A j = val ) 1 dv(D j :A j ) 10) However, ....

J. A. Blakeley, W. J. McKenna, and G. G. Graefe. Experiences building the Open OODB query optimizer. In Proc. ACM SIGMOD Conference, pp. 287--296, Washington, DC (1993).

....Toolkit (Alpha release from texas Instruments, Dallas [WBT92] as the underlying platform for this effort. We intend to concentrate on the open architecture and the issues mentioned above and plan on using available tools where possible. Our short immediate term goal is to use the Volcano [Mck92, BMG92] optimizer generator (by modifying it where necessary) to move towards the open architecture. Initially, we will try to group rules into modules that roughly correspond to local experts. This requires modifications to relevant parts of Volcano to match our requirements. As an intermediate term ....

Jose A. Blakeley, William J. Mckenna, and Gectz Graefe. Experiences building the open oodb query optimizer. Technical report, Texas Instruments INC., 1992.

.... memory) The materialization 1 Note that this blocking phenomenon arises even if operators are ones that support intra operator parallelism such the exchange operator of Volcano [CG94] 2 This notion of a materialization operator is not related to the operator for path expressions described in [BMG93] INRIA Dynamic Query Operator Scheduling for Wide Area Remote Access 7 operator provides tuples in response to next( requests from its parent operator when the parent is eventually able to execute. As stated in the introduction, there are three key policy questions for the implementation of a ....

J. Blakeley, W. McKenna, and G. Graefe. Experiences building the open OODB query optimizer. In Proc. of the ACM SIGMOD Int. Conf., page 287, Washington, DC, May 1993.

....version of the object algebra in typical object oriented databases to process spatial predicates as well as non spatial predicates [27] The operators in SOA are classified into algebraic operators and predicate operators. The algebraic operators are similar to the logical operators in Open OODB [6], MOOD [13] Paradise [17, 31] etc. They process inputs and outputs with a collection as a unit. The algebraic operators consist of the following: 1. relational algebra operators such as SELECT, PROJECT, JOIN, UNION, and DIFFERENCE, 2. NEST and UNNEST operators for set attributes in the extended ....

J.A. Blakeley, W.J. McKenna, and G. Graefe. Experiences building the open oodb query optimizer. In Proceedings of the ACM SIGMOD International Conference on Management of Data, pp. 287--296, Washington, D.C., ACM Press (1993).

....analogous to our mediator blades. Experimental database designs have gone beyond these more limited blade facilities to introduce far reaching extensibility. They have opened the entire database to modification through plug in modules. The Open OODB at Texas Instruments was one such project [5]. It enabled designers to change large granularity modules in order to introduce new or optimized functionality. The operating systems community has also experimented with component based approaches. Maybe the most famous of these is the Mach operating system [1, 26] It offered a kernel, ....

J.A. Blakeley, W.J. McKenna, and G. Graefe. Experiences building the open oodb query optimizer. In Proc. of the ACM SIGMOD Conf., 1993.

....Target object language, cost model, path expression, performance measurements. Cost Controlled OFL Rewriting Rules for Multiple Collection Traversals 2 Laboratoire PRiSM 1. Introduction The evaluation of path expressions is a crucial aspect in the optimization of objectoriented queries [Blakeley93, Kim95]. In fact, path expressions are the mechanisms most often used to query object bases [Carey88, Gardarin92, Kifer92] Path expressions describe multiple collection traversals navigations in the deeply nested structures of objects. Path expressions may be single valued or set valued, and may ....

....explored by the classic query rewriting and query planning phases of an optimizer is uniformly described by OFL programs. Query rewriting is concerned with finding equivalent expressions to the one given by the user [Haas90] In algebraic approaches, an algebraic expression represents a user query [Shaw90, Cluet92, Blakeley93]. Within the OFL formalism, an OFL program (a functional expression) represents the user query. Thus, equivalent expressions are equivalent OFL programs. Query planning explores the search space of execution plans to find the best one [Valduriez91, Lanzellote92, Mitchell93, Finance94] Within the ....

[Article contains additional citation context not shown here]

Blakeley J., Graefe G., "Experiences Building the Open OODB Query Optimizer", Intl. ACM SIGMOD Conf. on Management of Data, 1993.

....analogous to our mediator blades. Experimental database designs have gone beyond these more limited blade facilities to introduce far reaching extensibility. They have opened the entire database to modification through plug in modules. The Open OODB at Texas Instruments was one such project [6]. It enabled designers to change large granularity modules in order to introduce new or optimized functionality. The operating systems community has also experimented with component based approaches. Maybe the most famous of these is the Mach operating system [1, 24] It offered a kernel, ....

J.A. Blakeley, W.J. McKenna, and G. Graefe. Experiences building the open oodb query optimizer. In Proc. of the ACM SIGMOD Conf., 1993.

....to add new control strategies to the Volcano search engine, we had to introduce outside control over the Volcano, which in turn reduced the effectiveness of the approach. In this paper, we describe our experiences in developing the region based query optimizer through Cascades. The rule set used [BMG 93] is provided in the Appendix. These rules are sufficient to optimize both relational and object oriented queries. The performance comparisons between a region based query optimizer and an exhaustive query optimizer, both generated through Cascades, indicate that the region based optimizer has a ....

....guidance.region is star regionf find the subquery with the minimum cost modify memo( update memo structure such that the subquery with the minimum cost is joined with other subqueries g g return plan g 4. 2 Select Region In the complete rule set of the optimizer given in the Appendix [BMG 93] there are many rules for select classes (Rules 1,2,3,4,5,6,7,10,15,16) It has been observed that these transformations increase the optimization time a lot for queries with many select predicates. We have further noted that some of these rules can be disabled by using effective heuristics ....

Blakeley, J. A., McKenna, W. J., Graefe, G., "Experiences Building the Open OODB Query Optimizer ", Proc. of the ACM SIGMOD Conf., 1993.

....spatial predicates as well as non spatial predicates[21] 4 A simple oid list and an oid pair list are also kinds of an oid tuple list. The operators in SOA are classified into algebraic operators and predicate operators. The algebraic operators are similar to the logical operators in Open OODB[5], MOOD[13] Paradise[17, 24] etc. They process inputs and outputs with a collection as a unit. The algebraic operators consist of the following: 1) relational algebra operators such as SEL(ECT) PROJ(ECT) JOIN, UNION, and DIFF(ERENCE) 2) NEST and UNNEST operators for set attributes in the ....

J. A. Blakeley, W. J. McKenna and G. Graefe, "Experiences Building the Open OODB Query Optimizer," Proc. of ACM SIGMOD, 287-296, 1993.

....for existing OODBS, and a broad range of storage structures implemented on top of existing object servers. Algebraic Rewriting. Many optimization systems for object oriented query languages transform a query into an expression of a logical (object) algebra (e.g. VODAK ( AF95] Open OODB ( BMG93] as well as [CD92] Equivalences are defined on expressions of the algebra. The optimizer makes use of these equivalences by transforming expressions to find other algebraic representations of the query. Rule based optimizers heuristically prescribe direction and application of these ....

J.A. Blakeley, W.J. McKenna, and G. Graefe. Experiences Building the Open OODB Query Optimizer. In Proc. of the ACM SIGMOD Conf., pages 287--296, Washington DC, USA, May 1993.

....: A n ] for an extent belonging to object type E. It returns tuples consisting of the object identi er e and projects on the (possibly set valued) attributes A 1 ; A n . The algebraic counterpart of the idotj operator in OQL is the expand operator , also called, e.g. materialize [BMG93]. It may be used both to retrieve attributes and to invoke member functions of a referenced object. In this paper, we only need the attribute access variant (The operator ffi denotes tuple concatenation and g is a newly introduced attribute) g : e:a (E) fe ffi [g : e:a] j e 2 Eg Query 1: ....

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences building the Open OODB Query Optimizer. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 287295, Washington, DC, USA, May 1993.

....for object oriented query algebras and optimization by exploiting structural properties of complex datatypes is by now well known in the literature [5] 6] 7] 25] Also the use of the Volcano optimizer generator for query optimization in a structural oriented query algebra was investigated. In [8] an object oriented query algebra including a logical materialization operator for property paths was used. This operator can be considered as a predecessor of the operator map. The usage of Volcano for incorporating new data types and external methods in the context of scientific applications ....

J.A. Blakeley, W.J. Kenna, G. Gräfe: " Experiences Building the Open OODB Query Optimizer ", Proc. ACM SIGMOD International Conference on Management of Data, pp. 287--296, Washington, DC, May 26--28, 1993.

.... an extendible data model that is exploited in query processing [42] In the EXODUS VOLCANO projects generic support for query processing in extendible relational and objectoriented database systems was investigated and prototypes for query optimization and execution were developed [15] 21] 24] In [12] a query optimizer that was built on this prototypes for the Open OODB system is described. An extendible database programming language is described in [44] A proposal for modelling extendible search strategies of query optimizers is found in [36] Rule based query optimization is fundamental in ....

J.A. Blakeley, W.J. Kenna, G. Graefe: " Experiences Building the Open OODB Query Optimizer", preprint, Dec 1992.

....n ] for an extent belonging to object type E. It returns tuples consisting of the object identifier e and projects on the (possibly set valued) attributes A 1 ; A n . The algebraic counterpart of the dot operator in OQL is the expand operator c [KM90] also called, e.g. materialize [BMG93]. It may be used both to retrieve attributes and to invoke member functions of a referenced object. In this paper, we only need the attribute access variant (The operator ffi denotes tuple concatenation and g is a newly introduced attribute) c g : e:a (E) fe ffi [g : e:a] j e 2 Eg To flatten ....

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences building the Open OODB Query Optimizer. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 287--295, Washington, DC, USA, May 1993.

....simple puisqu il ne supporte que le balayage d une collection et la jointure de deux collections par boucles imbriqu ees, avec ou sans index. 7.0. 8 Volcano Volcano [GM93] est un g en erateur d optimiseurs de requetes, qui a servi de base notamment aux optimiseurs Cascades [Gra91] et OpenOODB [BMG93] Il est issu directement du g en erateur d optimiseur EXODUS [GD87] Volcano est un syst eme extensible avec lequel il est possible de programmer ses propres modules de r egles de transformation. Volcano supporte une alg ebre logique et une alg ebre physique, qui capture le mod ele d ex ecution. ....

J.A. Blakeley, W. McKenna, and G. Graefe. Experiences building the open OODB query optimizer. In Proc. ACM SIGMOD International Conference on Management of Data, Washington, DC, May 1993.

.... reduce operator can also be used to compute an aggregation (if we replace the [ annotation with an aggregate function, such as ) One of the most common optimization techniques for OODBs is to convert object references into joins [5] It is also known as the materialization of path expressions [1]. For example, the following nested OQL query: select distinct struct( E: e.name, M: e.manager.name ) from e in Employees can be converted into a pointer join, as shown in Figure 1.B. The join predicate compares the content of the attribute e.manager with the OID of an object m in the extent ....

....true ) x3, c, vars(e) and( x4, struct( bind(E,e) bind(M,x3) and( 4. 4 Translation of Paths into Joins One of the most common optimization technique for OODBs is converting object references into joins [5] It is also known as the materialization of path expressions [1]. Figure 3 summarizes these transformations. The rule in Figure 3.A indicates that if there is a unary operator op (such as the unnest operator) that depends on a path expression path 1 :path 2 , where path 1 and path 2 are paths (this dependency is denoted by a function f) and there is a class ....

J. Blakeley, W. McKenna, and G. Graefe. Experiences Building the Open OODB Query Optimizer. Proceedings of the ACM-SIGMOD International Conference on Management of Data, Washington, D.C., pp 287--296, May 1993.

....drafts of this paper. Notes 1. Note that this blocking phenomenon arises even if operators are ones that support intraoperator parallelism such the exchange operator of Volcano [10] 2. This notion of a materialization operator is not related to the operator for path expressions described in [5]. 3. In general, if n remote sources are subject to significant, independent delays, then by accessing those sources in parallel, scrambling has the potential to improve performance (over not scrambling) by as much as n times. 4. Thus, a query optimizer for a run time system that supports ....

J. Blakeley, W. McKenna, and G. Graefe. Experiences building the open OODB query optimizer. In Proc. of the ACM SIGMOD Int. Conf., page 287, Washington, DC, May 1993.

....of policy decisions that must be made with respect to the extent of scrambling: i) where in the tree to initiate scrambling; and ii) how many scrambling operations should be initiated. 2 This notion of a materialization operator is not related to the operator for path expressions described in [BMG93] For the first question, we consider two options: i) early initiation of a non leaf operator in the plan; and ii) early retrieval of data from a remote source. The first case, initiating a non leaf operator, requires the scrambling system to artificially call open( on that operator. The ....

J. Blakeley, W. McKenna, and G. Graefe. Experiences building the open OODB query optimizer. In Proc. of the ACM SIGMOD Int. Conf., page 287, Washington, DC, May 1993.

....of declarative This work was supported by the German Research Council under contracts DFG Ke 401 6 1 and SFB 346. queries. Albeit this optimization area has hitherto been somewhat neglected in many of the commercial systems, there are numerous proposals mostly from the academic world: [GD87, Bat86, BMG93, GM93] are proposals for optimizer generators; Fre87] has made rule based optimization popular in the relational context, which was adopted by the object oriented community [OS90, KM90, CD92] as well as by researchers in extended relational systems [BG92, Loh88] MDZ93, KMP93] proposed architectural ....

....operators listed above. The complete set of operators in our algebra also contains a join, an unnest, a projection, and a set difference operator. The function invocations and the attribute accesses are generalized to expansions an operator similar to the materialize operator as proposed in [BMG93]. The simplification step decomposes the query into its building blocks, and enriches them with statistical data, called basic values. For the purpose of the paper, we are mainly interested in selectivities and evaluation costs for function invocations. Among other things (explained below) ....

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences building the Open OODB Query Optimizer. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 287--295, Washington, USA, May 1993.

....rule implementations are not encapsulated. Consequently, conceptually simple modifications to a rule set often require significant effort including new function definitions to characterize the new feature. Modifying rule sets is actually quite important for systems like Starburst [15] Open OODB [5], or P2 [3, 4] where adding a new feature to a DBMS involves plugging a component into the DBMS itself. 1 For obvious reasons, updating a DBMS in such a manner should not require DBIs to hack rule sets in order to make the resulting DBMS execute correctly; rule sets should be automatically ....

.... different than Starburst and Volcano in using a building blocks methodology (as described in this paper) We have based our work on Volcano because it is freely available and because of Prairie s demonstrated usefulness [8] in re engineering large Volcano rule sets such as the Open OODB optimizer [5]. As described in [8] Prairie provides three key features that simplify the effort in writing rules. First, abstractions (like rules and actions) capture the design and semantics of an optimizer. This has the advantage that changes to an optimizer consists of changing the implementation of its ....

[Article contains additional citation context not shown here]

Jose A. Blakeley, William J. McKenna, and Goetz Graefe. Experiences building the Open OODB query optimizer. In Proceedings 1993 ACM SIGMOD International Conference on Management of Data, pages 287--296, Washington, May 1993.

....a set of reasonable execution plans for processing a given query, and selects the one with the minimum estimated cost. One of the most important factors that affects the cost estimation of an execution plan is the intermediate result cardinality the number of objects in an intermediate result [3, 7, 19, 21]. This is because the intermediate result becomes the operand of the next operation in the execution plan, and the cost of the next operation can be estimated mainly based on the number of objects in its operands [18, 19, 21] Hence, accurate estimation of intermediate result cardinality is a ....

....The selectivity of this nested predicate is the ratio of the number of path instances of the j D j A j D i A C i Fig. 4: Condition C i :A i [D j ] A j = val . path C i :A i [D j ] A j that satisfy A j = val to the total number of the path instances. Most of the conventional literature [2, 3, 9, 16, 17] on object oriented databases have no explicit estimation techniques for path selectivity. Kim et al. 11] use the technique proposed by Selinger et al. 19] for estimating the path selectivity as follows: sel(C i :A i [D j ] A j = val ) sel(D j :A j = val ) 1 dv(D j :A j ) 10) However, if ....

J. A. Blakeley, W. J. McKenna, and G. G. Graefe. Experiences building the Open OODB query optimizer. In Proc. ACM SIGMOD Conference, pp. 287--296, Washington, DC (1993).

....technique it will require the sequential scan of the extensions of the emp, dept, plant, and job classes, and even for very small sized extensions, this cost will exceed the cost of path traversal. On the other hand, the optimal execution plan for the following query is through explicit joins [Bla 93] SELECT emp.name, emp.job.name, emp.dept.name FROM Employee emp WHERE emp.dept.plant.location= Dallas It is clear that to be able to decide on the better execution plan, we should be able to calculate the cost of possible alternatives. Although the cost formulas for explicit joins and ....

.... techniques [Clu 92] and new execution algorithms to efficiently traverse complex object structures such as pointer based joins [She 90] and complex object assembly [Kel 91] Recently the design and implementation of a query optimizer based on complete extensible framework has been reported in [Bla 93] In [Bla 93] only the selectivities of the attributes are considered when there is index on them that can be used to assist selectivity estimation. In other cases, selectivities are assumed to be 10 . 2. Cost Model Parameters In the object model [Atk 92] used in this paper, complex objects ....

[Article contains additional citation context not shown here]

Blakeley, J. A., McKenna, W. J., Graefe, G., "Experiences Building the Open OODB Query Optimizer", Proc. of the ACM SIGMOD Conf., 1993.

....and translation managers for transferring objects among multiple address spaces, and a data dictionary providing name and type management facilities. The policy performer modules provide various database functionalities, one of which is query processing. Query processing in Open OODB [3] is largely influenced by the extensibility goals of the Open OODB project. The query module is an example of intra module extensibility. The query optimizer, built using the Volcano optimizer generator [13] is extensible with respect to algebraic operators, logical transformation rules, ....

....OAPT with a best subtree of Object Manager (OM) calls. These object manager calls are part of the set of low level object manipulation primitives that constitutes the interface to the OM. This approach respects the full encapsulation of objects. The methodology proposed in the Open OODB project [3] proposes to replace each logical algebra expression by an execution algorithm which is selected by the optimizer based on implementation rules. These rules use physical information that is encapsulated in physical property vectors (which is an abstract data type) This research proposes to ....

J.A. Blakeley, W.J. McKenna, and G. Graefe. Experiences building the Open OODB query optimizer. In Proc. ACM SIGMOD Int. Conf. on Management of Data, pages 287--296, 1993.

....as a mapping from logical algebraic terms that represent database queries into physical plans that represent scripts for evaluating the queries, such that the produced plans are optimal with respect to some cost model. This mapping can be captured in the form of a rule based term rewriting system [5, 6, 1]. The optimization process involves searching the space of equivalent program forms, generated mainly by the syntactic transformations captured as rewrite rules, and by the alternative access paths for accessing the same database objects, such as by the indices for retrieving data from relational ....

....while the expected order for the inner relation y is empty. That is, this rule propagates the expected order only to the outer relation of the join, because the order of the inner relation does not affect the sort order of the join result. Our experience with the Volcano optimizer generator [6, 1] revealed that there are some routinely performed coding tasks during an optimizer specification that are highly stereotyped and therefore amenable to automation. If they are not abstracted to system supported primitives, they consume a lot of energy during coding and they may result in obscure ....

[Article contains additional citation context not shown here]

J. Blakeley, W. McKenna, and G. Graefe. Experiences Building the Open OODB Query Optimizer. Proceedings of the ACM-SIGMOD International Conference on Management of Data, Washington, D.C., pp 287--296, May 1993.

....Algebra In this section, we describe how the base optimizer was extended to handle a more complex query algebra. The new algebra allows reference valued attributes, set valued attributes, and the use of path indices. We extended the optimizer to implement the optimization techniques described in [BMG93] We added a MATERIALIZE query algebra operator that represents materialization of a reference valued attribute (in other words, dereferencing a pointer) A corresponding ASSEMBLY algorithm class is used to represent the physical execution algorithm used to implement MATERIALIZE [KGM91] An ....

....sometimes evaluated using a path index without really having to materialize the individual components of the path expression. For example, if a pathindex exists on city.mayor.name, the predicate city.mayor.name = Lee can be evaluated without having to materialize the city or mayor objects (see [BMG93] for details) A new PATHINDEXSELECT algorithm was derived from the ALGORITHM class to capture such path index scans. A PATHINDEXSCANGENERATOR class was derived from the TREETOPLANGENERATOR class to replace occurrences of a string of materialize operators followed by a select operator in an ....

[Article contains additional citation context not shown here]

Jos'e A. Blakeley, William J. McKenna, and Goetz Graefe. "Experiences Building the Open OODB Query Optimizer". In Proceedings of the 1993 ACM-SIGMOD Conference, Washington, DC, May 1993.

.... next generation database systems [FMV93] Fre87, GD87, Loh88] made rule based query optimization popular, which was later adopted in the object oriented context, as e.g. OS90, KM90, CD92] Many researchers have worked on optimizer architectures that facilitate flexibility: Bat86, GD87, BMG93, GM93] are proposals for optimizer generators; HFLP89, BG92] describe extensible optimizers in the extended relational context; MDZ93, KMP93] propose architectural frameworks for query optimization in object bases. Besides these works on optimizer architectures, optimization strategies for both ....

.... It consists of the well known relational operators selection oe , projection , cartesian product Theta , join Gamma , semijoin Theta and Delta , division Xi , set difference Gamma , and union [ Furthermore, the algebra contains a so called expand operator comparable with [BMG93] s materialize operator or the map function [SS91] that is used for accessing object attributes, be they stored or computed (i.e. methods) For instance, the two path expressions f :to:country and f :distance( will be translated into to : f :to ; cou : to:country and dis : f ....

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences building the Open OODB Query Optimizer. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 287--295, Washington, DC, USA, May 1993.

....where the select predicate c.address.zipcode=44106 is the complex argument of the select operator which contains a path c:address:zipcode. Most existing object algebras [SZ90, KM94, CM94, CD92] are complex object algebras. One simple object algebra can be found in the TI OPEN system [BMG93]. Complex object algebras are good for users to communicate with the database; hence they are more suit Research partially supported by the NSF FAW award IRI 9024152 and the NSF grant IRI 92 24660. able for being used as query languages. However, they may not be good internal representations ....

....partially supported by the NSF FAW award IRI 9024152 and the NSF grant IRI 92 24660. able for being used as query languages. However, they may not be good internal representations for query processing, since it has been observed that it is cumbersome to process the expressions of complex algebras [BMG93]. One main reason is that it is not easy to find a small and efficient set of algebraic laws for complex object algebras, and hence the algebraic rewriting can not be done efficiently. On the other hand, simple object algebras have relatively more complete, but smaller set of algebraic laws. As ....

[Article contains additional citation context not shown here]

J.A.Blakeley, W.J.McKenna, and G.Graefe, "Experiences building the open OODB query optimizer," In proc. ACM SIGMOD International Conference on Management of Data, 1993.

....2 ( 43] and FAD( 6] that are discussed in section 5.1. 4.3 Query optimisation As the size of databases increases the need for efficient query algorithms becomes more apparent. Query optimisation can involve pre processing transformations on queries as well as efficient query execution plans. [12] and [42] present two examples of OO query optimisers. Query optimisation usually involves converting the query to some algebra that can then be manipulated to make it more efficient. This conversion process requires a detailed understanding of the semantics of the query processing and the data ....

Jos`e A. Blakeley. Experiences building the open OODB query optimizer. ACM SIGMOD Record, 22(2):287--296, June 1993.

....A query expressed in this high level format is parsed and transformed into an operator tree suitable for optimization. From this operator tree, the query optimizer generates an optimal access plan, which is then transformed into a C program ready for execution. The query optimizer in Open OODB [15] is generated using Volcano. It consists of 6 operators (SELECT, PROJECT, JOIN, RET, UNNEST (for set valued attributes) and MAT (MATerialize; it is used for representing path expressions in a query) 8 algorithms (Filter, Project, Hash join, Ptr hash join, File scan, Index scan, Unnest, and Alg ....

Jos e A. Blakeley, William J. McKenna, and Goetz Graefe. Experiences building the Open OODB query optimizer. In Proceedings 1993 ACMSIGMOD International Conference on Management of Data, pages 287--296, Washington, May 1993.

....and efficient processing of declarative queries. Fre87, GD87, Loh88] made rule based query optimization popular, which was later adopted in the object oriented context [OS90, KM90, CD92] Many researchers have worked on optimizer architectures that facilitate flexibility: GD87, Bat86, BMG93, GM93] are proposals for optimizer generators; BG92, HFLP89] described extensible optimizers in the extended relational context; MDZ93, KMP93] proposed architectural frameworks for query optimization in object bases. This work was supported by the German Research Council under contracts DFG ....

....operators listed above. The complete set of operators in our algebra also contains a join, an unnest, a projection, and a set difference operator. The function invocations and the attribute accesses are generalized to expansions an operator similar to the materialize operator as proposed in [BMG93] The simplification step decomposes the query into its building blocks, and enriches them with statistical data, called basic values. For the purpose of the paper, we are mainly interested in selectivities and evaluation costs for function invocations. Among other things (explained below) ....

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences building the Open OODB Query Optimizer. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 287--295, Washington, DC, USA, May 1993.

....as others to develop more rapidly new database query optimizers for noveldata models, query algebras, and database management systems. The Volcano optimizer generator has been used to develop optimizers for computations overscientific databases [WoG93] and for Texas Instruments sOpen OODB project [BMG93, WBT92], which introduces a new materialize or scope operator that captures the semantics of path expressions in a logical algebra expression. Both of these optimizers have recently become operational. Moreover, the Volcano optimizer generator is currently being evaluated by several academic and ....

J. A. Blakeley, W.J.McKenna, and G. Graefe, Experiences Building the Open OODB Query Optimizer, Proc. ACM SIGMOD Conf.,Washington, DC, May 1993, 287.

....as a globally known repository of system, object, name, and type information. Open OODB does not implement all its modules from scratch. The Exodus storage manager [CDR86] is used as an ASM for permanent storage of objects, and the Volcano optimizer generator to generate the query optimizer [BMG93]. For our purposes, the use of multiple threads, preferably on a multiprocessor platform, for event composition and rule firing in the active DBMS is essential. Therefore, we committed early to a Solaris 2.x platform. This required the porting of Exodus to the Solaris environment. Currently, a ....

Blakeley, J. A., McKenna, W. J., Graefe, G.; Experiences Building the Open OODB Query Optimizer. Proc. ACM SIGMOD, 1993.

No context found.

J. Blakeley, W. McKenna, and G. Graefe. Experiences building the Open OODB query optimizer. In Proc. ACM SIGMOD

No context found.

J. Blakeley, W. McKenna, and G. Graefe. Experiences building the Open OODB query optimizer. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 287--295, 1993.

No context found.

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences building the Open OODB query optimizer. In Proceedings of the ACM SIGMOD, pages 287--296, Washington, DC, USA, 1993.

No context found.

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences building the Open OODB query optimizer. In Proceedings of the ACM SIGMOD, pages 287--296, Washington, DC, USA, 1993.

No context found.

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences Building the Open OODB Query Optimizer. In Proceedings of ACM SIGMOD, pp. 287--296 (1993).

No context found.

J. Blakeley, W. McKenna, and G. Graefe. Experiences building the Open OODB query optimizer. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 287--295, 1993.

No context found.

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences building the open OODB query optimizer. In P. Buneman and S. Jajodia, editors, Proceedings of the 1993.

No context found.

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences building the open oodb query optimizer. In Proceedings of the SIGMOD Conference, pp. 287--296, 1993.

No context found.

J.A. Blakeley, W.J. McKenna, and G. Graefe. Experiences Building the Open OODB Query Optimizer. In Proc. ACM SIGMOD Conference, pp. 287--296, Washington, 1993.

No context found.

J. A. Blakeley, W. J. McKenna, and G. Graefe. Experiences building the open OODB query optimizer. In P. Buneman and S. Jajodia, editors, Proceedings of the 1993 ACM SIGMOD International Conference on Management of Data, pages 287--296. ACM Press, 1993. ftp://ftp.cs.pdx.edu/pub/ faculty/graefe/papers/OpenOODB.ps.

No context found.

Blakeley, J., McKenna, W., and Graefe, G. 1993. Experiences building the Open OODB query optimizer. In Proc. ACM SIGMOD Int. Conf. On Management of Data, 287--296.

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