Recent trends in memory sizes, combined with a demand for high-performance data management facilities, have led to the emergence of database support for managing memory-resident data as a topic of interest. In this paper we address the concurrency control problem for main memory database systems. Because such systems differ significantly from traditional database systems in terms of their cost characteristics, existing solutions to the problem are inappropriate; we present a new scheme based on two-phase locking that minimizes the overhead associated with concurrency control without overly limiting opportunities for concurrently executing transactions. We accomplish this by allowing the granularity of locking to vary dynamically in response to changes in the level of inter-transaction conflicts. Unlike hierarchical locking schemes, however, we avoid the expense of setting locks at multiple levels of a granularity hierarchy. We present a simple empirical analysis, based on instruction counts, to validate our claims.
Recently the relation of subsymbolic (“neural computing”) and symbolic computing has been a topic of intense discussion. Our purpose is to focus this discussion to the particular application area of expert system design. We address some of the drawbacks of current expert systems and study the possibility of using neural computing methodologies to improve their competence. The topic can be discussed at various levels of integration: the higher the integration level, the more symbolic functionalities (such as an inference engine) are implemented directly at the level of the neural computational model. In this paper we address the lowest levels of integration: neural networks that can be used to implement feature recognizers which allow symbolic inference engines to make direct use of complex sensory input via so called detector predicates. We also introduce the notion of self organization as a means to determine those attributes (properties) of data that reflect meaningful statistical relationships in the expert system input space, thus addressing the difficult problem of conceptual clustering (“abstraction”) of information. The concepts introduced are illustrated by two examples: an automatic inspection system for circuit packs and an expert system for respiratory and anesthesia monitoring. The adopted approach differs considerably from the earlier research on the use of neural networks as expert systems, where the only method to obtain knowledge is learning from training data. In our approach the synergy of rules and detector predicates combines the advantages of both worlds: it maintains the clarity of the rule-based knowledge representations at the higher reasoning levels without sacrificing the power of noise-tolerant pattern association (“inference by memory”) offered by neural computing methods.
They are two kinds of Statistical DataBases (SDBs), micro and macro-SDBs. Micro-SDBs describe individual entities or events, they are mainly used for statistical analysis. Macro-SDBs contain only summary-data achieved by statistical-mathematical operations on micro-SDBs. In this paper, we propose a data-structure called Complex Statistical Table (CST) which offer a formal and homogeneous framework to develop macro-SDBs. In order to manipulate CST structure, we define two classes of operators : CST transposition operators making data-organisation dynamic and CST retrieval operators that work on CST instances. We also propose a discussion pointing out our approach contrasted with other contributions of this field.
In this paper, two different structures for inverted files are analyzed and compared. The structures are called shared and separate inverted files. In the shared inverted file, the access keys can handle all the information usually handled by two separate inverted files. The results are given of some experiments which compare the shared structure with the separate one when the relational equi-join operation is executed. Moreover an evaluation is made of the memory space required by the shared and separate structures when a given application is taken into account.
Most previous approaches to incomplete information within the relational model depend on the specific semantics of the null types incorporated into this model. Herein we propose a model for incomplete information in nested relational databases which is independent of the semantics of the null types pertaining to incomplete information. Thus, the proposed model, called the nested relation type (NRT) model, allows user-defined null types, in addition to system-defined null types. The NRT model extends the nested relational model by incorporating a form of built-in inheritance. This allows us to define a partial order between nested-relations types and between the data values of these types. By utilizing these partial orders, we define an instance, over a NRT, to be incomplete when its information content may increase. In addition, we define an algebra for the NRT model, called the NRT algebra, which is shown to supercede known algebras for relations with nulls and for nested relations by showing faithfulness to these algebras. Finally, we investigate monotonicity of the operators of the NRT algebra, which allows us to predict how increasing or decreasing the information content of the instances in the database affects the user's view which is constructed from an algebraic expression over the instances in the database.
The exploration of spatial relationships is a multi-disciplinary effort involving researchers from linguistics, cognitive science, psychology, geography, cartography, semiology, computer science, surveying engineering, and mathematics. Terms like close and far or North and South are not as clearly understood as the standard relationships between integer numbers. The treatment of relationships among spatial objects is an essential task in geographic data processing and CAD/CAM. Spatial query languages, for example, must offer terms for spatial relationships; spatial database management systems need algorithms to determine relationships. Hence, a formal definition of spatial relationships is necessary to clarify the users' diverse understanding of spatial relationships and to actually deduce relationships among spatial objects. Based upon such formalisms, spatial reasoning and inference will be possible. The topological relationships are a specific subset of the large variety of spatial relationships. They are characterized by the property to be preserved under topological transformations, such as translation, rotation, and scaling. A model of topological relations is presented which is based upon fundamental concepts of algebraic topology in combination with set theory. Binary topological relationships may be defined in terms of the boundaries and interiors of the two objects to be compared. A formalism is developed which identifies 16 potential relationships. Prototypes are shown for the eight relationships that may exist between two objects of the same dimension embedded in the corresponding space.
Dynamic (virtual) hashing methods (DVH) manage very large files by the means of an index which is, at least partially, stored in the core. Usually this index is a binary tree (b-t), and different implementations of DVH are based on the representations of the b-t using pointers or links to descendants or leaves. We show that it is also possible to represent a p-ary tree (p-t), and thus a b-t, with a word of a certain language, and to use this pointerless representation to perform all the operations needed by a DVH generalized to p-t's. We present this language, we study its operative properties, and we indicate how to perform the operations of search, insertion,... We compare our representation to the most usual ones, and we analyze the complexity of some algorithms in relation with this representation.
The goal of the research on hashing as an access method is to arrive at a simple file organization scheme which guarantees single access retrieval from secondary storage, and is dynamic (can gracefully accommodate varying file sizes). There has been much progress in recent years in this regard. The Signature Hashing scheme for external files, proposed by Gonnet and Larson, guarantees single access retrieval [GL88]. But the scheme is static and implementation is complicated [LK84]. In this paper a simple external hashing scheme which guarantees single access retrieval is proposed and performance investigated. The scheme is dynamic, and its cost is better than that of Signature Hashing. An overview of the current state of the art and open problems are mentioned.
Timestamp ordering concurrency control mechanisms were considered to be quite suitable for distributed database systems, since transactions to be rolled badk can be determined locally at each site. Experiments, however, have shown that timestamp ordering mechanisms do not seem to be efficient and has a starvation problem for long transactions. In this paper, to improve efficiency of timestamp ordering mechanisms we propose to use a termination timestamp which is defined by a predicted commitment time or a predicted last read/write request time of a transaction. Besides other advantages the mechanism simplifies operations required for abort selection. The abort selection method introduced by the authors tries to improve the efficienty by selecting a proper transaction to be rolled back when conflict occurs. Comparison of several timestamp ordering methods obtained by combining these techniques is also given.
This thesis reports on joint research with Michael Rabin. Issues of operating system security occupy a central role in applied computer science; yet there has been no satisfactory complete solution to the problem of computer security. In the Security Toolkit Project we have developed a number of novel interlocking techniques which can be combined in many ways to provide tremendously enhanced security. The security toolkit, ITOSS, uses little overhead and is flexible; a security engineer can tailor a particular configuration to exactly satisfy the security needs demanded by the organizational structure at the site. It opens the way for further experimental work, rapid development, and simulation of new security schemes. The project consists of four phases: creating a new model of security, implementing the model by modifying UNIX (4.2 BSD) on a SUN-2 computer, inventing new algorithmic fences to validate the software by insuring that security violations are computationally infeasible, and developing software to permit easy exploitation of ITOSS’s features.
In this paper a new approach to locking multiversion composite objects is proposed, called the stamp locking approach. The main notion of this approach is a stamp lock defined as an extension of a classical lock in such a way that it contains the information about the position of locked nodes in the hierarchies concerned. The main advantage of this method is simple locking strategy following from the lack of intentional locks. To lock a hierarchy subtree it is sufficient to set one stamp lock only, whose node stamp identifies the subtree root. To determine stamp lock compatibility, it is sufficient to compare stamps that are particular identifiers of hierarchy nodes and subtrees. Advantages of the stamp locking method become particularly beneficial in the case of a database containing many versions of composite objects, whose composition structure is complex. This is a typical case of design databases where objects, being the artifacts of the design process, have usually many components available in many versions.
Advanced database systems have to support complex data structures as treated in object-oriented data models and nested relational data models. In particular, efficient processing of set-valued object retrieval (simply, set retrieval) is indispensable for such systems. In the previous paper [6], we proposed the use of signature files as efficient set retrieval facilities and showed their potential capabilities based on a disk page access cost model. Retrieval with signature files is always accompanied by mismatches called false drops, and it is very important in designing signature files to properly control the false drops. In this paper, we present an in-depth study of false drops in set retrieval with signature files. We derive formulas estimating false drops in four types of set retrieval based on the "has-subset," "is-subset," "has-intersection," and "is-equal" relationships. Then we evaluate their validity by computer simulations. Simulation study is also done to investigate false drops in practically probable more complex situations.
This paper describes Duma: a Data-based User interface management system for Multimedia Application, which is embedded in a multimedia document editor framework (MMDEF). MMDEF is the core of a multimedia document editor, which can adapt to externally defined media types and operations, and enables the user to work on documents composed of multimedia objects, including objects of newly defined types, through a coherent user interface. Duma introduces an extensible data model called interactor that abstracts the user interaction between application semantics and user interface components. Also, Duma's data-based UIMS architecture embodies an interactive UI design environment in which interfaces to the interactor model are given.
Filtered hashing is a new method of hashing that can maintain the benefits of hashing even when there are many overflow records. When an overflow occurs, a bucket is split into two and its address is stored in the overflow filter. When an underflow occurs, two buckets can be merged into one to reduce the file size. As long as the overflow filter fits in the main memory, the number of disk accesses to retrieve a record is guaranteed to be just one. The cost of inserting or deleting a record is also bounded by between 2 and 4 disk accesses. If the overflow filter grows too big due to excessive growth, the file can be reorganized with a bigger address space at a convenient time to achieve the optimal performance of static hashing, and can still adapt to dynamic changes later on.
This paper introduces the arc tree, a hierarchical data structure to represent arbitrary curved shapes. The arc tree is a balanced binary tree that represents a curve of length l such that any subtree whose root is on the k-th tree level is representing a subcurve of length l/2 k . Each tree level is associated with an approximation of the curve; lower levels correspond to approximations of higher resolution. Based on this hierarchy of detail, queries such as point search or intersection detection and computation can be solved in a hierarchical manner. We compare the arc tree to several related schemes and present the results of a practical performance analysis for various kinds of set and search operators. We also discuss several options to embed arc trees as complex objects in an extensible database management system and argue that the embedding as an abstract data type is most promising.
Multimedia computing requires support for heterogeneous data types with differing storage, communication and delivery requirements. Continuous media data types such as audio and video impose delivery requirements that are not satisfied by conventional physical storage organizations. In this paper we describe a physical organization for multimedia data based on the need to support the delivery of multiple playout sessions from a single rotating-disk storage device. Our model relates disk characteristics to the different media recording and playback rates and derives their storage pattern. This storage organization guarantees that as long as a multimedia delivery process is running, starvation will never occur. Furthermore, we derive bandwidth and buffer constraints for disk access and present an approach to minimize latencies for non-continuous media media stored on the same device. The analysis and numerical results indicate the feasibility of using conventional rotating magnetic disk storage devices to support multiple sessions for on-demand video applications.
Conventional Data Base Management Systems (DBMSs) are not generally effective for applications such as geographic data processing where data have spatial characteristics and queries involve on spatial relationships. These DBMSs can however be extended by supplementing them with special processing subsystems and new indexing structures and by augmenting the query interface language. DBMSs supporting an SQL interface are now widely used. The GEOgraphic Query Language (GEOQL) [18] is an extension of SQL proposed for geographic applications and supports both spatial and aspatial operations. In this paper, we propose a global optimization strategy for the hybrid queries so that a general query involving both spatial and aspatial selection can be executed efficiently. We show that the method is feasible.
