E. E. Chang and R. H. Katz. Exploiting inheritance and structure semantics for effective clustering and buffering in an object-oriented dbms. In Proceedings of the 1989.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....authors knowledge, the effect of clustering, a well researched OODB topic, upon XML and semi structured databases has never been investigated. Research into OODB clustering techniques has shown that effective clustering can improve the performance of some OODBs by up to several hundred percent [4]. Apart from the obvious effect on the number of disk accesses, clustering related objects together yields other performance improvements. For instance, better locality of reference can improve locking contention in highly concurrent systems. It could be expected that clustering would have a ....

....order, it is likely that the results of these queries will have to be materialized in document order. We propose that, in lieu of any further information, clustering an XML database in document ordering is a good default choice. recluster nodes in document order Order of reclustered nodes = [1, 2, 5, 10, 11, 3, 6, 7, 8, 12, 13, 14, 4] 3 4 8 7 6 5 10 12 13 14 11 1, 2, 5,10 11, 3, 6, 7 8, 12, 13, 14 4 1 2 3 4 Order of nodes on pages Figure 3: Static document order based clustering of XML database Figure 3 shows how a small database is rearranged in document order and how they are loaded on pages, assuming each page ....

E. E. Chang and R. H. Katz. Exploiting inheritance and structure semantics for effective clustering and buffering in an object-oriented dbms. In Proceedings of the 1989.

....algorithm ideally suited for mobile clients that use multi version concurrency control (MVCC) protocols [SS01] to relax conflicts between concurrent transactions. Since data caching is an effective (if not the most effective) and therefore indispensable way of reducing transaction response times [CK89] cache replacement policies have been extensively studied in conventional database management systems [EH84, OOW93, JS94, JN98, LKN 99] Since conventional caching techniques are inefficient for mobile networks where communication channels form an intermediate memory level between the client ....

E. Chang, R. Katz. Exploiting Inheritance and Structure Semantics for Effective Clustering and Buffering in an Object-Oriented DBMS. SIGMOD Conference 1989, pages 348-357.

.... about the type of use envisioned by a query execution plan, so that the system will know if re reference by the plan is likely and can act accordingly (see the Hot Set Model of [SACSCH] the DBMIN algorithm of [CHOUDEW] and its extensions [FNS, NFS, YUCORN] the hint passing approaches of [CHAKA, HAAS, ABG, JCL, COL] and the predictive approach of [PAZDO] This approach can work well in circumstances where re reference by the same plan is the main factor in buffering. In Example 1.2 above, we would presumably know enough to drop pages read in by sequential scans. The DBMIN algorithm would also deal well with ....

Ellis E. Chang, Randy H. Katz, Exploiting Inheritance and Structure Semantics for Effective Clustering and Buffering in an Object-Oriented DBMS, Proceedings of the 1989 ACM SIGMOD Conference, pp. 348-357. -17-

....algorithm ideally suited for mobile clients that use multi version concurrency control (MVCC) protocols [SS01] to relax conflicts between concurrent transactions. Since data caching is an effective (if not the most effective) and therefore indispensable way to reduce transaction response times [CK89] cache replacement policies have been extensively studied in conventional database management systems [EH84, CD85, SS86, RD90, JCL90, OOW93, JS94, JN98, LKN 99] Since conventional caching techniques are inefficient for mobile networks where communication channels form an intermediate memory ....

E. Chang and R. Katz. Exploiting Inheritance and Structure Semantics for Effective Clustering and Buffering in an Object-Oriented DBMS. SIGMOD Conference 1989, pages 348-357.

....new node. This is most effective when the arcs are inserted along with the node, that is, when a node is inserted, so are the arcs incident to it, although our method also supports insertion of isolated nodes, as described later. A similar heuristic was independently considered by Chang and Katz[15], who were interested in supporting clustering of objects incrementally in an CAD database system. The main difference between our approach and theirs is our more sophisticated page split algorithm, described next. Page splitting When an insertion of a node or an arc would render a page overfull, ....

E. Chang and R. Katz. Exploiting inheritance and structure semantics for effective clustering and buffering in an object-oriented database. In Proceedings of ACMSIGMOD 1989, pages 348--357. ACM Press, 1989.

....non procedural access data independence and these advances must not be compromised by third generation systems. Some argue that there are applications which never wish to run queries because of the simplicity of their DBMS accesses. CAD is often suggested as an example with this characteristic [CHAN89]. Therefore, some suggest that future systems will not require a query language and consequently do not need to subsume second generation systems. Several of the authors of this paper have talked to numerous CAD 4 application designers with an interest in databases, and all have specified a query ....

Chang, E. and Katz, R., "Exploiting Inheritance and Structure Semantics for Effective Clustering and Buffering in an Object-oriented DBMS," Proc. 1989 ACM-SIGMOD Conference on Management of Data, Portland, Ore., June 1989.

....is important to be able to enable and disable them. Enabling an optimization is essentially a performance hint to the underlying OMRDB. This general approach of application supplied hints has been adopted by other object to relational mapping systems, such as [20] and in other commercial products [8]. The main optimizations that are worth controlling in this way are the following: 1A. Attributes for 1 object When accessing an object O, prefetch all of O s scalar attributes. 1R. Relationships for 1 object When accessing an object O, prefetch all of O s references to other objects. MA. ....

Chang, E. E., Randy H. Katz, "Exploiting Inheritance and Structure Semantics for Effective Clustering and Buffering in an Object-Oriented DBMS," Proc. 1989 ACM SIGMOD Conf., pp. 348-357.

....has to be transparent from an user point of view. Furthermore, if the concept of cluster (via the notion of composite object) is part of the schema, when we want to modify the placement in secondary storage we have to perform a schema update Data independence is not insured anymore. In [CK89], it is shown that clustering exploits structural information such as composition hierarchy and inheritance properties. A simulation model has been developed which gives a comparative evaluation of different clustering strategies. The clustering algorithm studied in [CK89] chooses an initial ....

....not insured anymore. In [CK89] it is shown that clustering exploits structural information such as composition hierarchy and inheritance properties. A simulation model has been developed which gives a comparative evaluation of different clustering strategies. The clustering algorithm studied in [CK89] chooses an initial placement for newly created objects which depends on the most frequently used composition links between instances. The grouping strategies presented in the Encore system [ZH87] seem flexible and powerful (in particular the clustering by value feature) In Encore, the following ....

E. E. Chang, R. H. Katz, "Exploiting Inheritance and Structure Semantics for Effective Clustering and Buffering in an Object Oriented DBMS" in proc ACM SIGMOD conference Portland 1989.

....the read write ratio. Their method involves triggering a reorganization process to recluster the objects, which might not happen upon every insertion or deletion. They also assume knowledge of the traversal frequency in their leveled clustering method. The clustering scheme from Chang and Katz [10] requires information about traversal frequency on schema relationships. The algorithm makes a prioritized list of pages for insertion based on these frequencies. Most of these methods either assume a knowledge of database access patterns, or try to gather CHAPTER 3. HIERARCHICAL DATA CLUSTERING ....

E.E. Chang and R.H. Katz. Exploiting inheritance and structure semantics for effective clustering and buffering in an Obejct-Oriented DBMS. In Proceedings of ACM/SIGMOD Annual Conference on Management of Data, pages 348--357, 1989.

....DAGs. As another such example, a recent paper by Zou, Salzberg, and Ladin [ZSL97] describes a similar technique for clustering tree structured data. Other special purpose clustering mechanisms are designed to work with a specific type system or in a specific software context. Chang and Katz [CK89] describe a clustering mechanism that makes use of the structural and inheritance relationships of a CAD data model. Gruber and Amsaleg [GA94] describe a method that works in conjunction with garbage collection. Semantic clustering [SS90] is an interesting technique that combines clustering, ....

E. Chang and R. Katz. Exploiting Inheritance and Structure Semantics for Effective Clustering in an Object-Oriented Database System. In Proceedings ACM SIGMOD, pages 348--357, 1989.

....first. Most of the prefetching techniques found in the literature use a separate predictor component to determine the application s behavior in the near future. Predictors can either be strategy based [Jos70, Tri76, KE90, FP91] or trainingbased [BS76, Smi78, BC91, PZ91, CKV93] or structure based [CK89, KGM91, Gem91]. Strategy based predictors incorporate an explicit programmed strategy whereas training based predictors require extensive monitoring for consolidating the statistics during run time. Future references of an application are predicted by exploiting the reference patterns obtained in the past. ....

....sequences of pages whereas in the last years some techniques were proposed for database systems which try to predict sequences of objects . The following table shows the related work according to the classification scheme: Unit of prediction Prediction mechanism Objects Pages Predictor based [CK89, KGM91] [Jos70, BS76, Tri76, Smi78] BC91, Gem91, PZ91] FP91, KE90, CKV93] Code based [WZ86, KL91, CKP91] RL92, CB92, MLG92] UNIX madvise Predictor Based Prefetching Techniques. One essential characteristic of predictor based prefetching techniques is the time of prefetching. Basically, there exist two ....

[Article contains additional citation context not shown here]

E. E. Chang and R. H. Katz. Exploiting inheritance and structure semantics for effective clustering and buffering in an object-oriented DBMS. In Proc. of the ACM SIGMOD Conf. on Management of Data, pages 348--357, Portland, OR, May 1989.

....Lee and Smith [11] designed a branch prediction table which is a small associative memory that retains the addresses of recently executed branches and their targets. Structure based predictors (mostly used in OODBMSs) obtain information from the object structure. Chang and Katz s technique [3] predicts the future access from the data semantics in terms of inheritance and structural relationships, e.g. configuration and version history. An Assembly operator for complex objects to load sub objects in advance was introduced by Keller [10] The traversal was performed by different ....

.... advance was introduced by Keller [10] The traversal was performed by different scheduling algorithms (depth first and breadth first) In object oriented databases the unit of I O is an object (Object Server) or a page (Page Server) An object server prefetches an object or a group of objects ([3], 4] and a page server prefetches one or more pages ( 5] 7] Another possible classification of prefetching is the time factor. Smith [16] proposed two policies: a) prefetch only when a buffer fault occurred (demand prefetch) b) prefetch at any time (prefetch always) Prediction ....

E. E. Chang and R. H. Katz. Exploiting Inheritance and Structure Semantics for Effective Clustering and Buffering in an Object-Oriented DBMS. In Proceedings of the ACM SIGMOD Conference on the Management of Data (Portland, Oregon, May-June 1989).

....analysis is used to extract data access semantics and a cost model helps evaluate the benefits of the initial clustering strategy and dynamically adapt it. There has also been research on the use of semantics of the object model (such as inheritance and structure semantics) for smart clustering [5]. Semantic clustering [13] has been proposed as a method for achieving good quality clusters at lower costs, provided that data access semantics can be captured and specified effectively. A more recent trend in clustering research has been to use graph partitioning algorithms for clustering the ....

E. E. Chang and R. H. Katz, "Exploiting Inheritance and Structure Semantics for Effective Clustering and Buffering in an Object-Oriented DBMS", Proc. of the ACM SIGMOD Intl. Conf. on Management of Data, Portland, OR, 1989, pp. 348-357.

....of access, every edge will have the same weight in their stochastic model. The clustering technique of Cheng and Hurson [CH91] also uses weighted graphs, and suffers from the same problem as above when applied to access structures. Other work on clustering in object oriented databases includes [CK89, LS92]. None of these addresses the issue of worst case path lengths for access structures. 4 Tree Structures In this section, we assume the access structure is a tree. We consider the case when the access structure is a DAG later. Section 4.1 addresses the static case, where the tree is given and its ....

Ellis E. Chang and Randy H. Katz. Exploiting inheritance and structure semantics for effective clustering and buffering in an object oriented dbms. In Procs. of the ACM SIGMOD Conf. on Management of Data, pages 348--357, May 1989.

....page faults during the subsequent use of those objects. The roots of the techniques used are typically found in the techniques developed for packing records into pages [60, 75, 59, 49] A significant amount of work has also been done on the specific problems of clustering objects into pages[6, 31, 67, 62, 15, 28]. Fetching a multi object page instead of a single object is the only prefetching done in most object systems. Accordingly, for dynamic prefetching to be of any value, it has to do something better than can be done by clustering objects into pages. Clustering can be seen as a particularly static, ....

....group computation in Thor can happen either statically or dynamically, as described in this chapter. As was shown by the sparse and dense traversals, very different computations sharing the same objects can benefit from dynamic prefetching. There has been some work arguing for dynamic reclustering [15, 50]; however, such reclustering is expensive compared to dynamically computing prefetch groups (since reclustering involves disk writes) and it is challenging to decide when it is worth reclustering. A problem common to both clustering and dynamic prefetching is that it can be difficult to tune the ....

[Article contains additional citation context not shown here]

Ellis E. Chang and Randy H. Katz. Exploiting inheritance and structure semantics for effective clustering and buffering in an object-oriented dbms. In Proceedings of the ACM SIGMOD Conference on Management of Data, pages 348--357, 1989.

....pages sequentially from disk. The IPC cost could be reduced by using a disk thread instead of a disk process. 1.2 Approaches to Improve Database Performance In order to reduce the high cost of a page fetch many researchers have developed prefetching techniques. In objectoriented databases Chang [3] predicts future accesses by hints from the data semantics in terms of inheritance and structural relationships to prefetch one object ahead. Cheng [4] extended this work by adding multiple hints and by taking into account the prefetching depth and physical storage considerations. A complex ....

E.E. Chang and R.H. Katz. ExploitingInheritance and Structure Semantics for Effective Clustering and Buffering in an Object-Oriented DBMS. In Proc. of the ACM SIGMOD Int. Conf. on the Management of Data, pages 348-- 357, Portland, Oregon, June 1989.

....prefetching strategy where all objects of a buffered page are copied into the object buffer upon the first occurence of an object fault against that page. More sophisticated prefetching policies, based either on semantic knowledge about the structure and operation on complex object structures [11], or on profiling information [28] have also been reported. Flushing the entire object cache at the end of a transaction, even augmented with good prefetching policies, is not applicable to PJama for two reasons. First, because all threads concurrently access the object cache and their ....

E.E. Chang and R. Katz. Exploiting inheritance and structure semantics for effective clustering and buffering in an oject-oriented dbms. In Proc. of the ACM SIGMOD Int. Conf. on Management of Data, May 1989.

.... queries [Sto91, ML97, Sa95, Jo98] Prefetching in database systems has been studied mostly for applications where future access patterns are largely predictable due to specific structures of the underlying databases and the programs accessing them, especially in objectoriented database systems [CK89, CH91, GK94], but also in real time and multimedia applications [WZ86, MKK95, TP97] The effect of object prefetching is implicitly achieved (on a per page basis) also by intelligently clustering objects into pages [CK89, TN91, TN92, GKKM93] In the field of mobile computing, KP97] has pursued a similar ....

.... the programs accessing them, especially in objectoriented database systems [CK89, CH91, GK94] but also in real time and multimedia applications [WZ86, MKK95, TP97] The effect of object prefetching is implicitly achieved (on a per page basis) also by intelligently clustering objects into pages [CK89, TN91, TN92, GKKM93]. In the field of mobile computing, KP97] has pursued a similar approach where reference distances in user and program access patterns are used for predictive file hoarding. 144 There is only little work on prefetching based on probabilistic models. Fundamental properties of Markov chainbased ....

Chang EE, Katz RH (1989) Exploiting Inheritance and Structure Semantics for Effective Clustering and Buffering in an Object-Oriented DBMS. In: ACM SIGMOD Conf., 1989, Portland, Ore., pp 348--357

.... work in clustering looks at grouping objects logically using information about the object s structural (component) relationships or also inheritance relationships [16, 25, 131] Recent research also addresses the dynamic re organization of object clusters as structural object relationships change [27]. Related research dynamically determines prefetch units using pattern access histories [113] Cluet and Delobel [31] use clustering and index information to limit the search for equivalent queries during query rewrite. Lanzelotte et al. 91] represent clustering and index information 2 The ....

Ellis E. Chang and Randy H. Katz. Exploiting Inheritance and Structure Semantics for Effective clustering and Buffering in an Object-Oriented DBMS. In SIGMOD Proceedings, pages 348--357. ACM, 1989.

....guarantees of a separate protected server domain. Our techniques are effective when server access is read mostly, that is, the majority of RPC calls read but do not modify server data, while a minority of calls change that data. For example, some CAD applications exhibit read mostly behaviour [3]. A Dexter based hypermedia system [4] also needs mostly read only access to the the data for each node in the hypermedia document. In our scheme, a server exports certain data segments read only to clients. This allows some client requests to be satisfied by reading data directly from ....

Ellis E. Chang and Randy H. Katz. Exploiting inheritance and structure semantics for e#ective clustering and bu#ering in an object-oriented DBMS. In ACM International Conference on Management of Data, 1989.

....objects within a cached page can be marked as missing. Both schemes studied in the thesis use object level marking. 1. 2 Database and Workload Assumptions In addition to assuming that the normal case for an OODB is to have multiple objects per database page, we assume that a clustering algorithm [3, 14, 19, 51, 54, 55] is used to produce a well clustered database with respect to common access patterns. In other words, a transaction accessing page P is likely to access multiple P objects. The database can be very large (much larger than the client or server caches) We assume the set of pages accessed by the ....

.... Clustering and Write Probabilities We assume that the objects stored in the database have been placed onto pages by some clustering process that uses information about common access patterns or the links between objects to determine groups of objects that are likely to be accessed together [3, 14, 19, 51, 54, 55]. Clustering is important for performance; the objects on a page are read from disk as a group and are sent to a client as a group. We model the quality of clustering by specifying a range of cluster sizes. The average cluster size determines the average number of objects accessed per page. To ....

E. Chang and R. Katz. Exploiting inheritance and structure semantics for effective clustering and buffering in an object-oriented DBMS. In ACM SIGMOD International Conference on Management of Data, pages 348--357, Portland, OR, June 1989.

....to let the query optimizer tell the buffer manager the plan of the query to be processed, so that the buffer manager can allocate and manage its buffers accordingly. Algorithms include the Hot Set model [18] the DBMIN algorithm [5] and related extensions [9, 13, 23] and hint passing algorithms [4, 12, 1, 10, 3]. In [15] O Neil, O Neil and Weikum point out that it is difficult to design a query plan based buffering algorithm that works well in a multitasking system. Sequences of transactions, concurrent transactions, and partions of the current transaction plan can overlap in complicated ways. In ....

E.E. Chang and R.H. Katz. Exploiting inheritance and structure semantics for effective clustering and buffering in an object-oriented dbms. In Proc. 1989 ACM SIGMOD Conf., pages 348-- 357, 1989.

....an object access model upon a cluster, the dynamic reclustering, a simulation model for performance measurement, a physical data store model for a cluster, and a model for access frequency measurement. Several papers published on the clustering of object oriented databases discussed these issues [1, 3, 4, 12]. However, they assumed that there were already the partition blocks of related objects. For example, in ORION system[8] instances of the same class are located in the same physical segment. Furthermore, instances which belong to user specified collection of classes are located in the same ....

E.E. Chang, and R.H. Katz, "Exploiting Inheritance and Structure Semantics for Effective Clustering and Buffering in an Object-oriented DBMS," Proc. of ACM SIGMOD Conference, pp. 348-357, 1989.

No context found.

E. Chang and R. Katz, "Exploiting Inheritance and Structure Semantics for Effective Clustering and Buffering in an Object-Oriented DBMS," Proceedings of the ACM SIGMOD Conference, p. 348 (May-June 1989).

No context found.

Chang, E.E. and Katz, R.H., "Exploiting Inheritance and Structure Semantics for Effective Clustering and Buffering in an Object-Oriented DBMS," Proceedings of ACM SIGMOD International Conference on Management of Data, 1989. Discusses dynamic clustering based on structural links (object to object relationships) and inheritance.

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