Ante Vilenica

Benchmarking and Evaluation Support for Self-Adaptive Distributed Systems

Increasingly, distributed systems have to deal with highly dynamic and hardly predictable environments. This trend, in conjunction with rising demands for sophisticated non-functional system requirements, challenges both the development and operation (i.e. management) of traditional distributed systems. One promising approach to cope with these challenges are self-adaptive distributed systems that are characterized by the capability to configure and maintain themselves. However, the inherent dynamic of self-adaptive systems requires intensive evaluation and benchmarking efforts in order to ensure the intended system behaviour. In order to support that, this paper presents a framework that aims at supporting the nominal-actual comparison of self-adaptive distributed systems as well as the comparison of different self-adaptive solutions with respect to a specific software implementation task. The underlying approach consists of (i) a declarative definition language and (ii) a software component that is capable of conducting evaluations and benchmarks on different software implementations.

A Reputation-Based Approach to Self-Adaptive Service Selection

Service-orientation provides concepts and tools for flexible composition and management of largescale distributed software applications. The automated run-time management of such loosely coupled software systems, however, poses still major challenges and is therefore an active research area, including the use of novel computing paradigms. In this context, the dynamic and adaptive selection of best possible service providers is an important task, which can be addressed by an appropriate middleware layer that allows considering different service quality aspects when managing the adaptive execution of distributed service workflows dynamically. In such an approach, service consumers are enabled to delegate the adaptive selection of service providers at run-time to the execution infrastructure. The selection criteria used are based on the cost of a service provision and the continuous, dynamic evaluation of reputations of providers, i.e. maintained track records of meeting the respective service commitments. This paper discusses the design and operating principle of such an automatic service selection middleware extension. Its ability to balance different quality criteria for service selection, such as service cost vs. the reliability of provision, is empirically evaluated based on a multi-agent platform approach.

Simulation Management for Agent-Based Distributed Systems

The development of open distributed applications with autonomous components is increasingly challenged by rising complexity caused, e.g., by heterogeneous software and hardware modules as well as varying interactions between such components. For such systems, simulation provides a well-established way to study the effects of different system configurations on its behavior as, e.g., especially its performance. Therefore, simulation is of particular interest for autonomous distributed applications, i.e. for those that exhibit emergent phenomena and where simulation is required for, e.g., validation purposes within the development process. Prominent application domains targeted at in this work include the areas of multi-agent and self-organizing systems. For easing and speeding up simulations of system characteristics when developing such systems, this work presents tools and a powerful software framework for automating the corresponding simulation management. Its basis is a declarative language that is used to describe the setting to be simulated as well as its evaluation. Then, an additional software framework is provided that processes this description automatically - thereby relieving the developer from manually managing the execution, observation, optimization and evaluation of the simulation. In summary, this leads to much easier and sounder system simulations and, thus, to developing better autonomous distributed systems and applications.

Using Building Blocks for Pattern-Based Simulation of Self-organising Systems

The constantly rising complexity of distributed systems and an increasing demand for non-functional requirements lead to approaches featuring self-organising characteristics. Developing these systems is challenged by their hardly predictable dynamics and emergent phenomena and requires therefore the incorporation of simulation techniques. In doing so, not all needed development activities can be realised by just one software application because self-organisation often implies unique settings, goals, and development methods as well as the use of individual code sections. In order to handle such unique environments, this contribution presents a pattern-based concept that incorporates reusable patterns for different development issues of self-organising systems by encapsulating various methods, algorithms, and applications in so called building blocks and combining them in a coherent and hierarchical process.

Methods and Tools for Engineering Self-Organizing Software Systems

Developing software systems that can cope with constantly changing contexts and at runtime is a challenging task. Most importantly, changing contexts require adaptive system behavior in order to maintain system functionality even under dynamically changing environment conditions. And this relates not only to the core system functionality but also to non-functional aspects such as, e.g., reliability and/or scalability. In order to deal with such challenges adequately, this paper presents an approach that adapts the paradigm of Self-Organization to computer software systems. It enables systems to preserve their main caracteristics while changing their structure in consequence to outside influences without the existence of any central controller. So, the specific aim of this paper is to address the utilization of Self-Organization in engineering software systems. Accordingly, it presents a set of methods and tools as developed in a recent research project for engineering self-organizing software systems. The underlying approach is organised around state-of-the-art software engineering phases such as system modelling, programming, and the respective software development procedure.

Abstract User Interfaces for Mobile Processes

An important focus of recent business process management systems is on the distributed, self-contained and even disconnected execution of processes involving mobile devices. Such an execution context leads to the class of mobile processes which are able to migrate between mobile and stationary devices in order to share functionalities and resources provided by the entire (mobile) environment. However, both the description and the execution of tasks which involve interactions of mobile users still require the executing device and its context to be known in advance in order to come up with a suitable user interface. Since this seems not appropriate for such decentralized and highly dynamic mobile processes, this work focuses on the integration of manual tasks on the respective ad-hoc creation of user interfaces at runtime. As an important prerequisite for that, this paper first presents an abstract and modality-independent interaction model to support the development and execution of user-centric mobile processes. Furthermore, the paper describes a prototype implementation for a corresponding system infrastructure component based on a service-oriented execution module, and, finally, shows its integration into the DEMAC (Distributed Environment for Mobility-Aware Computing) middleware.

Multi-Agent-Architecture for simulating traffic management — A case study on highway networks

This paper addresses the automated management and simulation of capacity problems for motor highway infrastructures, which often cannot keep up with traffic increases produced by, e.g., cars, motorbikes and trucks. Such kinds of problems require rather intelligent highway infrastructure management with strategies with, often, contradictory goals: On the one hand side, individual drivers would like to use a traffic infrastructure which suits their individual needs best, e.g. with respect to time, resources, and personal driving comfort, and, on the other, the highway operators may want to optimize the use of their respective traffic infrastructure w.r.t. security, cost, and overall user satisfaction. This paper claims that such kinds of problems can be modelled best with using agent-oriented models and techniques. Therefore, it demonstrates how to use a generic multi-agent-system architecture that supports both modelling and simulating of such an application. It demonstrates how such an architecture can be used for applying, analyzing and comparing different traffic management strategies, inspired by different sources like network management, feedback control, or nature inspired coordination algorithms.