Songchun Moon

Global query management in heterogeneous distributed database systems

Abstract Global query management provides the ability to combine data from different local databases in a single retrieval operation. The necessity for global query management arises in an open, heterogeneous multidatabase system, since autonomy and heterogeneity of component databases have given rise to a number of new major issues regarding the global query optimization strategy and context mediation including data conversion and query translation. For instance, some local DBMSs never support semijoin operator which has been proposed in order that data transmission between sites could be reduced. In this regard, the global query optimization strategies developed for homogeneous distributed database systems make extensive use of semijoins which are not attractive in the multidatabase context since this may increase the local processing time. Moreover, these do not consider the cost incurred as a result of data conversion and query translation. Our basic principle to get a high performance is that we decompose a global query to the finest level of subqueries in order to explore all possible execution plans. This reduces the overall processing cost in an heterogeneous distributed database system because subqueries that are sent to the local databases have to be converted. The lesser number of subqueries implies less conversion cost.

Formalized Entity Extraction Methodology for Changeable Business Requirements

Without a formal methodology extracting entities from business descriptions, a business requirement in the real world cannot be abstracted correctly into an entity-relationship schema. Once core entities are discovered, we can obtain an Entity-Relationship Diagram (ERD) by inserting relationships between/among the relevant entities and by aggregating some attributes into one of the entities or relationships. There have been so many studies on formal guidelines for extracting entities from business descriptions. Most of them adopt a knowledge-based approach which consults a knowledge base to recommend entity candidates. However, the knowledge-based approach usually fails to construct the most appropriate ERD for a given business domain. The approach performs the entity extraction on the stiff premise that an object would be classified as an entity if it happen to be classified as an entity once or more in past applications. The previous studies did not consider the flexibility in the object classification that even the same object could be regarded as either an entity or an attribute according to the various concerns of field workers. In this paper, we discuss some limitations of the previous researches on object classification and propose a new methodology for flexible entity extraction. To evaluate the practicality of the devised methodology, we developed a tool for the methodology and performed a case study on option trading applications with the tool.

An efficient log-based crash recovery scheme for nested transactions

In this paper, an efficient recovery scheme for nested transactions using double log sequence numbers(LSNs) is proposed, Unlike previous recovery schemes, due to double LSNs, our recovery scheme is able to avoid unnecessary redos and subsequent undos. Furthermore, our scheme can cope with successive crashes which might occur during recovery. In addition, our scheme provides operation logging as well as value logging, which is required in order to allow semantically-rich lock modes. The recovery scheme based on the selective redo/undo consists of three passes: analysis, redo, and undo.

Resilient data management for replicated mobile database systems

We propose a new replication control scheme for mobile database systems. Replicating data inevitably induces the burden of maintaining consistency, which requires more complex synchronization mechanisms, especially in the presence of communication failures. If a communication link fails, all sites are divided into two disjoint groups: the major group called the quorum partition and the minor group called the non-quorum partition. This phenomenon, called network partitioning, could lead to access starvation. In order to access replicated databases without access starvation even in the presence of network partitioning, we propose a new scheme called fault-tolerant quorum consensus (FTQC). FTQC is based on the idea that quorum formation is allowed even in the non-quorum partition. FTQC guarantees inter-partition one-copy serializability, without sacrificing data availability.

A new indexing scheme supporting multi-attribute database applications: MAX

Abstract Multi-attribute indexing schemes can be classified into seven classes according to the manner of partitioning multi-di-mensional data space into regions, each of which contains all the records of a single data page. On the basis of this principle, we classify according to three properties of the hyperplane partitioning a region: the dimension of hyperplane, the number of hyperplanes, and the normal vector of hyperplane. Among the seven classes, we select a class as our indexing scheme model according to the complexity for maintaining hyperplane. From our model, we derive an indexing scheme, MAX , which handles multi-attribute data efficiently. In addition, a number of algorithms for manipulating multi-attribute data are given, together with their computational and I/O complexity. Moreover, we show that MAX is a kind of generalized B-tree. This means that MAX can be easily implemented on existing built-in B-trees in most storage managers in the sense that the structure of MAX is like that of B-tree. We measure the performance by testing against various realistic data and query sets. Results from the benchmark test indicate that MAX outperforms other indexing schemes on insertion , range query , and spatial join . For insertion, together with B-tree using z-transformation , MAX presents good performance in the aspects of CPU and I/O cost. Regardless of various clustering factors, the storage cost of MAX is remarkably low compared with KDB-tree and R-tree. We conclude that MAX is an efficient indexing scheme for multi-attribute database. If MAX is fully implemented as an index method of storage manager, the performance of many applications using multi-attribute will be remarkably enhanced.

Efficient storage management for large dynamic objects

In this paper, we propose a new database storage manager, called Buddy-size Segment Storage System (BSSS), to handle large dynamic objects of any size and then implement search, insertion, append, and deletion algorithms used for the storage structure. The internal nodes of the proposed storage manager are identical to the ones proposed in Exodus. However, unlike Exodus that has fixed-size segments for leaf blocks, BSSS has binary buddy-size leaf segments. The proposed storage manager is compared with Exodus through performance results from simulation approach. BSSS performs the same as or better than the best case of Exodus for object creation time, sequential scan time, and random search time. However, the insertion performance of BSSS is the same as or worse than the one of Exodus.<<ETX>>

Commit-reordering validation scheme for transaction scheduling in client-server based teleputing systems: COREV

In this thesis, we propose new validation schemes, COREV and R*COREV. for scheduling transactions optimistically in client-serverbased mobile database systems. In the mobile database system environment, validation throughput could be seriously degraded due to waiting time in validation process resulting from various delay factors in data transfer through wireless channels. Unlike the previous approaches, our schemes remove the waiting time by rearranging commit order of transactions. R*COREV is devised to reduce abort ratio by reanaaging the serializable order of them. For preserving serialimbility corresponding to the substantial execution order, which could be perverted easily in mobile environments, we devised a new scheme for assigning transaction number, called setialorder restoration scheme for transaction numbering (SORT). It enables transactions to be executed in a serializable order and validated without any waiting time.

Prototyping DIME, a tightly coupled heterogeneous distributed database system

In this paper, we design and implement a prototypical heterogeneous distributed database system, named DIME (Distributed Information Management). DIME has the following salient properties. First, DIME allows both global retrieval operations and global update operations where different concurrency control schemes are used in different local database systems (LDBSs). Second, DIME implements international standard protocols on the distributed transaction processing and the remote database access. Last, DIME provides distribution transparency, and thus users can generate not only single site queries (including remote site queries) but also inter-site queries without considering data distribution in LDBSs.<<ETX>>

Concurrent View Maintenance Scheme for Soft Real-Time Data Warehouse Systems

The data warehouse should be refreshed periodically or aperiodically in order to prevent OLAP transactions from reading extremely stale data. This refreshing process is referred to as view maintenance. There have been many researches on ways to preserve data consistency during the process of view maintenance. The simplest approach among them is to perform view maintenance in the nighttime while OLAP transactions are allowed to be active only in daytime. However, this approach cannot be applied to globalized corporations because no common idle time is available for such corporations. In this paper, we propose a new algorithm for performing view maintenance without blocking OLAP transactions. Our algorithm improves data freshness by allowing one or more maintenance transactions and OLAP transactions to be executed concurrently. We introduce the details of algorithm with some examples, present proofs of devised scheme, and then evaluate performance gains by means of simulation method.

Lightweight multigranularity locking for transaction management in XML database systems

As eXtensible Markup Language (XML) provides a capability for describing data structures, and can accommodate many kinds of semistructured data. The semistructured data format is flexible for changing data structures through insertion and deletion of data elements in mission-critical applications. In the case of concurrently changing such a data format, this flexibility could be endangered by a phantom problem which might lead to inconsistent information flow. For the purpose of developing a concurrency control scheme without the phantom phenomenon, we propose a lightweight multigranularity locking (LWMGL) scheme that is a hybrid mechanism of Tree-based Locking and Multigranularity Locking. The goal of this scheme is to realize locking at the level of precise elements in an XML database while preventing the phantom problems. Since these precise locks could considerably reduce the number of pseudo-conflicts that are regarded as unnecessary locks, they provide high concurrency compared with other concurrency control schemes. In order to realize the LWMGL scheme we also devised a new data model of XML indexed element tables (XIETs) for transferring diverse XML documents. This data model does not only can preserve the XML tree structure in application levels, but also enables execution of the structural change operations as well as the data access operations in parallel.