Search

The efficient processing of database applications by computing systems with multi-tiered persistent storage devices needs specialized algorithms to generate optimal persistent storage management plans. A correct allocation and deallocation of multi-tiered persistent storage may significantly impact the overall performance of data processing in such systems. This paper describes the new algorithms, that create persistent storage allocation plans for data processing on multi-tiered persistent storage devices. This work is based on the pipelined data processing model, and it applies a formalism of extended Petri nets to describe the data flows between the tiers of persistent storage devices. The algorithms presented in the paper perform linearization of the extended Petri nets and generate the optimal persistent storage allocation plans. The paper describes the experiments conducted to validate the generate data allocation plans for multi-tiered persistent storage devices and prove the improvements in performance compared with random data allocation plans.

It is well known that analysis of database workloads recorded in the past can be used to improve performance of a database system in the future. This work investigates a problem how to use information extracted from the audit trails to predict the characteristics and levels of future workloads in a database system. The paper describes a new approach to a database workload forecasting based on the identification of periodically repeated instances of events and association of workload related information with the events. The proposed approach creates the new workload forecasts from the analysis of audit trails in the periods of times related to the earlier determined instances of events. A system of operations on the workload forecasts defined in this work can be used to derive the new workload forecasts from the already existing ones. An event based approach to workload forecasting that associates the specific types of database workloads with the instances of events allows for more precise the forecasting of such workloads in the future data processing.

Performance-based quality evaluation of database conceptual schemas is an important aspect of database design process. Performance of user applications strongly depends on the logical and physical database structures derived from the conceptual schemas. This work presents a complete process of performance-based quality evaluation of conceptual schemas. It shows how to express a conceptual database schema in a notation of simplified UML classes. It proposes a new specification of object algebra for representation of conceptual level database applications. Transformation of conceptual schemas and expression of object algebra into implementation schema described in the paper allows for precise estimation of the processing costs of database applications and for precise evaluation of performance-based quality of conceptual schemas. An experiment conducted as a proof of concept for the performance-based quality evaluation procedure concludes the paper.

Apart from providing rich data security information, database audit trails are the good sources of information on the workload characteristics of database system. This is why the periodic patterns of database operations found in the audit trails can be used for both performance tuning and intrusion detection. This work defines a new concept of an irregular periodic pattern and it shows how such patterns can be detected in the audit trails of database operations. The paper presents the algorithms that implement a new way of finding the irregular periodic patterns in audit data. The quality and complexity of the algorithms together with the outcomes from the experiments are discussed in the paper.

An online integration system enables incremental computation shortly after an increment data arrived at the central site. Processing increments serially ensures all data containers are in their updated states for computation of the next increment data. In general, a data container may show up as several arguments in a data integration expression. Serial processing of increments at this data container failed to show its best performance due to expensive IO costs for materialization updates. This paper proposes an online integration system with dynamic scheduling to enable concurrent processing of increments of data. The online integration system allows a series of transformation of a data integration expression into a single increment expression upon the increments of multiple data containers, and generates a data integration plan. The dynamic scheduling system employs a monitoring system and a priority scheduling which is able to dynamically change the data integration plans according to the increment data behavior.

Information about periodic computations of processes, events, and software components can be used to improve performance of software systems. This work investigates mining periodic patterns of events from historical information related to processes, events, and software components. We introduce a concept of a nested event log that generalizes historical information stored in the application traces, event logs and dynamic profiles. We show how a nested event log can be compressed into a reduced event table and later on converted into a workload histogram suitable for mining periodic patterns of events. The paper defines a concept of periodic pattern and its validation in a workload histogram. We propose two algorithms for mining periodic patterns and we define the quality indicators for the patterns found. We show, that a system of operations on periodic patterns introduced in this work can be used to derive new periodic patterns with some of the quality indicators better from the original ones. The paper is concluded with an algorithm for deriving periodic patterns with the given quality constraints.

One of the distinguishing characteristics of a data stream system is ?a continuous query operating on dynamic data? as opposed to ?static data processed by instantaneous query? in DBMS. Due to potentially infinite size of the stream a query is evaluated after forming finite subsets of the data stream. The sliding window model is the most suitable model for processing finite subsets of a data stream where only recent data items qualify. In this paper, we propose a mathematical model to express a continuous query on sliding windows. This model expresses a window as an ordered set. Its sliding mechanism is expressed through a set of transition operations and query evaluation is expressed through a set of output operations.

This work applies some of the techniques used in mining of association rules in large relational databases to fusion of data from large spatial databases. The specific objective is to show that algorithms developed for the identification of frequent itemsets in the relational databases can be, after slight modifications, successfully used to fuse complex two dimensional objects from large amounts of range data collected with ultrasonic sensors. The paper presents two algorithms that use this approach to construct polylines and polygons from ultrasonic range data.

The article addresses a problem of optimal query processing in heterogeneous and distributed multidatabase systems. The objective is to find a method for optimization of global query processing plans. Two generic strategies of query processing in heterogeneous and distributed systems include entirely sequential and entirely parallel approaches. The paper describes a hybrid strategy where some of the subqueries are submitted and processed sequentially while the others are processed in parallel. Optimization of query postprocessing is proposed and described in order to reduce an impact of errors in estimation of processing costs at the local systems.