Dependency analysis is important in all of the stages of workflow processes. An appropriate analysis will help us to identify the potentially affected entities if changes occur. In this technical report we thoroughly analyze the critical entities of workflow processes and their dependency relationships, and propose a multi-dimensional dependency model, that includes routing dependency, data dependency and role dependency. We further build a knowledge base to represent the multi-dimensional dependency model using the logic programming language Prolog. We then develop a set of query rules that can be applied to the well-defined knowledge base at both activity and process levels to retrieve the potentially affected entities. Finally we use a case study of workflow processes in the healthcare domain to show how our dependency analysis methodology works.

Research on software agents has produced a diversity of conceptual models for high-level abstract descriptions of multi-agent systems (MASs). However, it is still difficult and costly for designers that need a unique set of agent modeling features to either develop a new agent modeling language from scratch or undertake the task of modifying an existing language. In addition to the modeling itself, in both cases a significant effort need to be expended in building or adapting tools to support the language. An extensible agent modeling language is crucial to experimenting with and building tools for novel modeling constructs that arise from evolving research. Existing approaches typically support a basic set of modeling constructs very well, but adapt to others poorly. A declarative language such as XML and its supporting tools provides an ideal platform upon which to develop and extensible modeling language for multi-agent systems. In this paper we describe xTAO, an extensible agent modeling language, and also demonstrate its value in the context of a real-world application.

In this work we show that adaptive page refresh models can be further improved by careful selection of initial page-refresh rates of newly added links as indicated by our page evolution studies showing that page-change rates (and consequently page-refresh rates) are dependent on top-level domain, category and page depth.

In general, current structural joins cannot exploit views efficiently when the view definition is not a prefix (or a suffix) of the XPath query. To increase the applicability of our proposed view definition, we propose a novel physical structural join operator called InterJoin. The InterJoin operator allows for joining interleaving XPath expressions, e.g., joining //A//C with //B to evaluate //A//B//C. InterJoin treats structural joins as a logical operator, giving more join alternatives in XML query plan.

We propose several physical implementations for InterJoin, including a technique to exploit spatial indexes on the inputs. We give analytic cost models for the implementations so they can be costed in an existing join-based XML query optimizer. Experiments on real and synthetic XML data show significant speed-ups of up to 200% using InterJoin, as well as speed-ups of up to 400% using our materialized views.

Two aspects of the retrieval system -- fast, incremental updates and garbage collection for delayed document deletions -- are discussed in detail, with a focus on the respective trade-offs. Depending on the relative number of queries and update operations, different strategies lead to optimal overall performance. Special attention is given to a particular case of dynamic search systems -- desktop and file system search. As one of the main results of this paper, we demonstrate how security mechanisms necessary for multi-user support can be extended to realize efficient document deletions.

This paper discusses several ways to arrange postings in memory and studies the effects on memory requirements and indexing performance. Starting from the traditional approach that holds all postings for one term in a linked list, we examine strategies for combining consecutive postings into posting groups and arranging these groups in a linked list in order to reduce the number of pointers in the linked list. We then extend these techniques to compressed posting lists and finally evaluate the effects they have on overall indexing performance for both static and dynamic text collections. Substantial improvements are achieved over the initial approach.

Data partitioning is often used to scale-up a database system. In a centralized database system, the serialization order of commited update transactions can be inferred from the database log. To achieve this in a shared-nothing distributed database, the serialization order of update transactions must be inferred from multiple database logs. We describe a technique to generate a single stream of updates from logs of multiple database systems. This single stream represents a valid serialization order of update transactions at the sites over which the database is partitioned.