Alfons Kemper, Christoph Kilger, and Guido Moerkotte. Function materialization in object bases. In Proceedings of the 1991.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....violate abstraction and encapsulation. They require knowledge of the implementation of a type to name the path of interest, constrain types to one representation, and require that the key values of interest be represented directly in the type s implementation. Method precomputation schemes[8, 10, 35, 34] allow queries based on method results, but they do not support indexes based on other functions and the only sets that can be indexed are type extents, sets that contains all the objects of a particular type. 1.2. Distributed Databases 15 1.2 Distributed Databases We expect object oriented ....

....that point that does the right recomputation using the new value, but trying to determine the effect of a particular change to the result of an index function is likely to require a user to write the needed function. This type of scheme would be similar to the compensating actions in the GOM system[34, 35] that we will see later in this chapter. In addition, our object model allows multiple representations of a type, thus the object graph for a function may not be identical for each element in the indexed set; we would need to provide an additional group of index data parts for each distinct graph ....

[Article contains additional citation context not shown here]

Alfons Kemper, Christoph Kilger, and Guido Moerkotte. Function materialization in object bases. In Proceedings of the 1991.

....violate abstraction and encapsulation. They require knowledge of the implementation of a type to name the path of interest, constrain types to one representation, and require that the key values of interest be represented directly in the type s implementation. Method precomputation schemes[8, 10, 35, 34] allow queries based on method results, but they do not support indexes based on other functions and the only sets that can be indexed are type extents, sets that contains all the objects of a particular type. 1.2. Distributed Databases 15 1.2 Distributed Databases We expect object oriented ....

....that point that does the right recomputation using the new value, but trying to determine the effect of a particular change to the result of an index function is likely to require a user to write the needed function. This type of scheme would be similar to the compensating actions in the GOM system[34, 35] that we will see later in this chapter. In addition, our object model allows multiple representations of a type, thus the object graph for a function may not be identical for each element in the indexed set; we would need to provide an additional group of index data parts for each distinct graph ....

[Article contains additional citation context not shown here]

Alfons Kemper, Christoph Kilger, and Guido Moerkotte. Function materialization in object bases. Technical Report 28/90, Universitat Karlsruhe, October 1990.

....object types (for example, Print method on objects, Area over polygons, etc. form a serious obstacle for optimal plan generation, since their costs are not directly available to the query optimizer. A variety of strategies for handling this situation have been proposed in the literature [KKM91, GCD 93] In BODHI, we have extended the ODL language to allow optional de nition of cost functions, and functionally equivalent methods. These extensions enable the cost based optimizer to compute the cost associated with each of the equivalent methods, before choosing the best execution ....

A. Kemper, C. Kilger, and G. Moerkotte. Function Materialization in Object bases. In SIGMOD

....as paths of instance variables stored in objects. There are several problems with this approach. First, it does not support keys derived from multiple instance variables. Second, evaluating keys and noticing modifications requires help from the compiler. A few database systems, such as GOM [18], provide index maintenance using function materialization. This is similar to functional indexes in TDB, except GOM maintains and stores the materialized keys with the object, while TDB computes them when objects are opened for writing, which reduces space usage and may reduce computation. ....

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. In Proceedings of the ACM SIGMOD International Conference on Management of Data, 1991.

....view definition for object oriented databases [6,7] made us aware of the great demand for effective optimization of such queries. Optimization techniques reported recently include to estimate the cost of method invocation for cost based query optimizers[2,8] and the precomputation of method calls [9,10,11] following the lines of view materialization. The essence of the latter approach is to convert computed data (i.e. the result of method calls) to stored data so that all access structures and optimization techniques for stored data can be employed. Obviously, this technique requires considerable ....

Kemper A., Kilger C., Moerkotte G. Function Materialization in Object Bases. SIGMOD Proceedings, ACM,

....structures are constructed and materialized on demand. This may imply partial construction of view objects (e.g. delayed evaluation of virtual attributes) In addition, the rematerialization overhead incurred by updates to a base can be reduced by applying techniques such as those described in [KKM91], where the characteristics of the object oriented paradigm (encapsulation and object identity) are largely exploited. Virtual Object Storage In the presence of materialized views, some storage strategies can be defined to reduce retrieval time for virtual structures. The system can, for ....

Alfons Kemper, Christoph Kilger, and Guido Moerkotte. Function Materialization in Object Bases. In Proc. ACM SIGMOD Conference on Management of Data, pages 258--267, 1991.

....virtual classes formed using selection queries on aggregation paths (or short, path query views) To the best of our knowledge, only two other research groups have addressed this topic. Kemper et al. s work on function materialization addresses the problem of precomputing function results [6, 7]. Konomi et al. discuss a solution to supporting a type of join class that is formed along the aggregation graph [8] Traditional indexing techniques are not well suited for this task of maintenance. The indexing needs of the path query view problem are unique in that because the contents of the ....

....class. They do not address the more general problem of path query views, which is the focus of this paper, nor do they provide any cost models or performance analyses. Function Materialization. The work of Kemper et al. on function materialization is closely related to OODB view materialization [6, 7]. The goal of function materialization is the precomputation and maintenance of function results. Similar to the SMX satisfaction indicator solution, Kemper et al. associate a validity value with each object that can serve as an argument to a function. However, our satisfaction indicators ....

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. SIGMOD, pages 258--267, 1991.

....class. They do not address the more general problem of path query views, which is the focus of this paper, nor do they provide any cost models or performance analyses. Function Materialization. The work of Kemper et al. on function materialization is closely related to OODB view materialization [5, 6]. The goal of function materialization is the precomputation and maintenance of function results. Similar to the SMX satisfaction indicator solution, Kemper et al. associate a validity value with each object that can serve as an argument to a function. However, our satisfaction indicators ....

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. SIGMOD, pages 258--267, 1991.

....virtual classes formed using selection queries on aggregation paths (or short, path query views) To the best of our knowledge, only two other research groups have addressed this topic. Kemper et al. s work on function materialization addresses the problem of precomputing function results [7, 8]. Konomi et al. discuss a solution to supporting a type of join class that is formed along the aggregation graph [9] Traditional indexing techniques are not well suited for this task of maintenance. The indexing needs of the path query view problem are unique in that because the contents of the ....

....class. They do not address the more general problem of path query views, which is the focus of this paper, nor do they provide any cost models or performance analyses. Function Materialization. The work of Kemper et al. on function materialization is closely related to OODB view materialization [7, 8]. The goal of function materialization is the precomputation and maintenance of function results. Similar to the SMX satisfaction indicator solution, Kemper et al. associate a validity value with each object that can serve as an argument to a function. However, our satisfaction indicators ....

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. SIGMOD, pages 258--267, 1991.

....called materialized functions have been investigated in [KMWZ91] The functions are used for deriving attribute values which are stored in a separate data structure. A similar strategy is pursued in the extended relational database system Postgres [SK91] Cost models and benchmark results are in [KKM91] and [Han87] for relational databases) Object oriented Query Optimization. Object oriented database systems like O 2 [OT92, BCD92] and ObjectStore [OHMS92] focus query optimization on the use of physical clustering strategies and indexes. In O 2 , indexes are quite flexible by allowing a ....

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. In Proc. 1991 ACM SIGMOD Int. Conf. on Management of Data, pages 258--267, Denver, Colorado, May 1991.

....predicate migration (Hellerstein 1994) tackles the problem in a practical fashion but leaves space for more optimization, especially of more complex join predicates. Another approach to handle user defined functions is their materialization, i.e. storing the results of the function invocation (Kemper, Kilger, and Moerkotte 1991; Bertino 1991) Chaudhuri and Shim (1993) investigate the problem from the point of rewriting. Similar to our approach they assume that a set of rules specifying the semantics is available to the optimizer. 3 Optimizing Queries With Foreign Functions The typical way to use foreign functions in a ....

Kemper, A., C. Kilger, and G. Moerkotte (1991). Function materialization in object bases. In Proc. ACM SIGMOD Conference on Management of Data, pp. 258--268.

....are now used to compute for every expression e with weight w e the set of weighted path expressions P(e) occurring in e. 5 The formal definition of the decapsulation process is given in Appendix A. The decapsulation process presented here is a refinement of the decapsulation process applied in [KKM91]. If e is some constant P(e) or if e is some variable v , P(e) f v[0] g. If e is of the form e = v:A 1 : Delta Delta Delta :A k P(e) f v[w e ] A 1 [w e ] Delta Delta Delta : A k Gamma1 [w e ] A k [0] g Here, all weights are set to w e except for the last one that is set to ....

....the above fashion, variables are replaced by their corresponding types, and all resulting sets of weighted path expressions are merged to one set with weights of matching expressions being accumulated. Index structures like access support relations [KM90] or generalized materialization relations [KKM91] have a significant impact on clustering. If there is an access support relation, i.e. an index supporting a path expression p of the operation op, then p should not be considered for clustering and thus its weights have to be set to 0. If the results of the operation op are materialized and ....

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. In Proc. of the ACM SIGMOD Intl. Conf. on Management of Data, pages 258--267, Denver, CO, May 1991.

....tree with execution information. The query tree, when fully expanded, will be input to a rule based optimizer such as an EXODUS optimizer[10] Kemper, Kelger and Moerkotte explored the materialization of function results and the requirements for updating these results as an object base is modified [15]. They propose to incorporate the materialized functions into a rule based optimizer. Each area of research attacks a specific problem in object based query optimization and proposes a new technique for solving that problem. In the next section we describe an architecture that will allow the ....

A. Kemper, C. Kilger, and G. Moerkotte, "Function Materialization in Object Bases," in SIGMOD Proceedings, pp. 258--267, ACM, 1991.

....problem in optimization of object oriented expressions is path expressions. Such expressions imply a navigation through objects to find the end of a path. Research in this area includes defining indexes for paths [19, 79, 97, 111] optimization in the presence of arbitrary methods along the path [20, 60, 62, 78], and the use of clustering and other storage information to determine path accesses [31, 75, 91] An alternative approach is to define a complete system for optimization. The application of algebraic transformations has formed the basis for the design of many query optimizers for objectoriented ....

....are placed to allow the system to detect when a precomputed method is invalidated. Precomputed results, when still valid, are retrieved using the index. Invalidated results require the method be computed at query execution time. A similar approach is taken by Kemper, Kilger and Moerkotte [78]. In this system, precomputed function results are stored in relations (called materialized functions) along with validity flags for the data, and information about the objects contributing to a function result is maintained. The properties of types, encapsulation and object identity are used to ....

[Article contains additional citation context not shown here]

Alfons Kemper, Christoph Kilger, and Guido Moerkotte. Function Materialization in Object Bases. In SIGMOD Proceedings, pages 258--267. ACM, 1991.

....structures are constructed and materialized on demand. This may imply partial construction of view objects (e.g. delayed evaluation of virtual attributes) In addition, the rematerialization overhead incurred by updates to a base can be reduced by applying techniques such as those described in [KKM91], where the characteristics of the object oriented paradigm (encapsulation and object identity) are largely exploited. Virtual Object Storage In the presence of materialized views, some storage strategies can be defined to reduce retrieval time for virtual structures. The system can, for instance, ....

Alfons Kemper, Christoph Kilger, and Guido Moerkotte. Function Materialization in Object Bases. In Proc. ACM SIGMOD Conference on Management of Data, pages 258--267, 1991.

....(or short, path query views) To the best of our knowledge, only two other research groups have addressed the topic of maintaining materialized path query views in object oriented databases. Kemper et al. s work on function materialization addresses the problem of precomputing function results [7, 8]. Konomi et al. discuss a solution to supporting a type of join class that is formed along the aggregation graph [10] Readers might also note that in [12] we discussed the path query view problem and proposed an initial technique for the maintenance of such views; however, none of the techniques ....

....class. They do not address the more general problem of path query views, which is the focus of this paper, nor do they provide any cost models or performance analyses. Function Materialization. The work of Kemper et al. on function materialization is closely related to OODB view materialization [7, 8]. The goal of function materialization is the precomputation and maintenance of function results. Similar to the SMX satisfaction indicator solution, Kemper et al. associate a validity value with each object that can serve as an argument to a function. However, our satisfaction indicators ....

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. SIGMOD, pages 258--267, 1991.

....Furthermore, while traditional path indices are traversed primarily in a forward direction in order to answer queries, materialized path query views use the index structures for propagating updates, and hence, they traverse the index structures backwards. Similarly, function materialization [7] and path query view materialization are closely related, because function materialization uses the index structures to store the materialized functions results. However, function materialization also depends on indices for queries, with the propagation of updates an auxiliary (if unavoidable) ....

....path query view materialization are closely related, because function materialization uses the index structures to store the materialized functions results. However, function materialization also depends on indices for queries, with the propagation of updates an auxiliary (if unavoidable) concern [7]. In this paper, we present our work towards the support of more powerful view mechanisms, namely, virtual select classes defined using path queries, e.g. queries that are formed using aggregation paths. We first identify a number of limitations of traditional index structures. We then propose a ....

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. SIGMOD, pages 258--267, 1991.

....called materialized functions have been investigated in [KMWZ91] The functions are used for deriving attribute values which are stored in a separate data structure. A similar strategy is pursued in the extended relational database system Postgres [SK91] Cost models and benchmark resuls are in [KKM91] and [Han87] for relational databases) There are several other systems combining advantages of reasoning in concept languages with object oriented (e.g. SLT91, LRS92] or rule based (e.g. BHHM91] representation languages by a hybrid system architecture. The idea to use different concept ....

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. In Proc. 1991 ACM SIGMOD Int. Conf. on Management of Data, pages 258--267, Denver, Colorado, May 1991.

....is the same basic idea that is being pursued in the revelation project [GM88] Decapsulation opens many paths to optimization in object bases, where the mostly procedural approach makes algebraic optimization difficult. First steps in this direction were made by Kemper, Kilger and Moerkotte in [KKM91], where decapsulation is applied to determine accessed attributes in the context of function materialization, and by DeWitt and Lieuwen in [LD91] in their work about loop optimization. To estimate the cost for evaluating a function, it is necessary to use an adequate cost model. In Section 4.2.1 ....

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. In Proc. of the ACM SIGMOD International Conference on Management of Data, pages 258--268, 1991.

....up to date. In the following paragraphs, we shall summarize the basic concepts that ensure the integrity of materialized function results. It must be stressed that there exist several refinements in order to keep the rematerialization expenses to a minimum; details can be derived from [13] [14] and [15] The Generalized Materialization Relation (GMR for short) is the data structure that holds materialized function results. In its simplest form, it contains the result for all possible parameter combinations. Therefore, only non atomic arguments (i.e. object types) are permitted in this ....

A. Kemper, C. Kilger, and G. Moerkotte, "Function materialization in object bases," in Proc. of the ACM SIGMOD International Conference on Management of Data, pp. 258--268, 1991.

....case of good clustering, cannot be tolerated. In this paper, we propose a new page oriented predictor called Prefetch Support Relation (PSR) which combines the advantages of both techniques mentioned above. PSRs are very similar to the Generalized Materialization Relations (GMRs) introduced in [11, 12] which are used to store materialized (i.e. precomputed) function results. GMRs store the precomputed function result and are, therefore, only applicable for side effect free functions. On the other hand, PSRs facilitate support even for operations which do cause side effects, i.e. operations ....

.... Delta Delta] Page 5: id 17 ; id18 ; Delta Delta Delta ; id88 ] Page 6: id 26 ; id27 ; Delta Delta Delta] Page 7: id 28 ; Delta Delta Delta] a) Objects (b) Object Placement Figure 2: Sample Database with Storage Structure we call this the materialization of the function volume [11]. All function results are stored in a separate index structure called GMR. In this case we only need to read one page, namely the page containing the materialized results of function volume . However, materialization is only applicable for side effect free functions. Now, consider the second ....

[Article contains additional citation context not shown here]

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 258--268, Denver, USA, May 1991.

....kemper informatik.rwth aachen.de [kilgerjmoer] ira.uka.de 3 This work was partially supported by the German Research Council DFG under contract number SFB 346 and Ke 401 6 1. An excerpt presenting selected issues appeared in: Proc. ACM SIGMOD Conf. on Management of Data, Denver, CO, May 1991 [10]. Abstract View materialization is a well known optimization technique of relational database systems. In this work we present a similar, yet more powerful optimization concept for object oriented data models: function materialization. Exploiting the object oriented paradigm namely ....

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 258--267, Denver, CO, May 1991.

....the relational model over more than two decades. 4. Many queries naturally return relations and not objects (see also [27] on this) 5. Allowing formal reasoning with new (relation based) optimization techniques, e.g. access support relations [16, 17] and generalized materialization relations [15]. For us, these arguments seemed cogent enough to incorporate the relational model into our formal model of object oriented databases. 2.1 Sorts Within this subsection we start specifying the value part of our model. We assume that the set of attribute symbols Attr : fA 1 ; A 2 ; g and the ....

....the query is (entirely) based on extensions and returns a relation, the knowledge of relational query processing and optimization can be utilized. But still, some object oriented specialities must be considered. First, access structures especially developed for object oriented databases (e.g. [15, 17]) must be taken into account. Second, the cost model for estimating the overhead of different evaluation alternatives must be adapted to the particularities of the object oriented model. Furthermore, since object oriented data models have more expressive power than the relationals the optimization ....

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 258--268, Denver, USA, May 1991.

.... The Index Structures The GOM query evaluation is supported by two very general index structures tailored for object oriented data models: ffl Access Support Relations (ASRs) 12] are used to materialize (frequently) traversed reference chains, and ffl Generalized Materialization Relations (GMRs) [11] maintain pre computed function results. Since these two index structures have to be taken into account in the optimization process, two index relations based on the schema Company are exemplified: Emp:worksIn:mgr] #0 : OIDEmp #1 : OIDDept #2 : OIDManager id1 id5 id8 id2 id5 id8 : ....

A. Kemper, C. Kilger, and G. Moerkotte. Function materialization in object bases. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 258-- 268, Denver, USA, May 1991.

No context found.

A. Kemper, C. Kilger and G. Moerkotte, Function Materialization in Object Bases, Proc. ACM SIGMOD Int. Conf. Management of Data, pp.258--267, 1991.

First 50 documents

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