## Fachbereich Informatik

- SDL Design Patterns and Components - Watchdog and Heartbeat (2004)
- We present a methodology to augment system safety step-by-step and illustrate the approach by the definition of reusable solutions for the detection of fail-silent nodes - a watchdog and a heartbeat. These solutions can be added to real-time system designs, to protect against certain types of system failures. We use SDL as a system design language for the development of distributed systems, including real-time systems.

- Ein maßgeschneidertes Kommunikationssystem für eine mobile Applikation mit Dienstgüteanforderungen (2004)
- In diesem Beitrag wird die Maßschneiderung eines Ad-Hoc-Kommunikationssystems zur Fernsteuerung eines Luftschiffs über WLAN vorgestellt. Dabei steht die Dienstunterstützung bei der Übertragung mehrerer Datenströme im Vordergrund. Es werden verschiedene Dienstgütemechanismen erklärt und deren Entwicklung und Integration in ein Kommunikationsprotokoll mit Hilfe eines komponentenbasierten Ansatzes genauer erläutert.

- Induction of Linear Temporal Logic Programs (2004)
- We propose a framework for the synthesis of temporal logic programs which are formulated in a simple temporal logic programming language from both positive and negative examples. First we will prove that results from the theory of first order inductive logic programming carry over to the domain of temporal logic. After this we will show how programs formulated in the presented language can be generalized or specialized in order to satisfy the specification induced by the sets of examples.

- On the Efficiency of Multiple Importance Sampling (2004)
- Approximating illumination by point light sources, as done in many professional applications, suffers from the problem of the weak singularity: Numerical exceptions caused by the division by the squared distance between the point light source and the point to be illuminated must be avoided. Multiple importance sampling overcomes these problems by combining multiple sampling techniques by weights. Such a set of weights is called a heuristic. So far the estimators resulting from a heuristic only have been analyzed for variance. Since the cost of sampling is not at all constant for different sampling techniques, it is possible to find more efficient heuristics, even though they may hove higher variance. Based on our new stratification heuristic, we present a robust and unbiased global illumination algorithm. By numerical examples, we show that it is more efficient than previous heuristics. The algorithm is as simple as a path tracer, but elegantly avoids the problem of the weak singularity.

- Compositional Testing of Communication Systems (2004)
- Today, test methods for communication protocols assume, among other things, that the protocol design is specified as a single, monolithic finite state machine (FSM). From this specification, test suites that are capable of detecting output and/or transfer faults in the protocol implementation are derived. Limited applicability ofthese methods is mainly because oftheir specific assumptions, and due to the size of the derived test suite and the resulting test effort for realistic protocols. In this work, the compositional test method (C-method), which exploits the available structure of a communication protocol, is proposed. The C-method first tests each protocol component separately for output and/or transfer faults, using one of the traditional test methods, then checks for composability, and finally tests the composite system for composition faults. To check for composability and to derive the test suite for the detection of composition faults, it is not required to construct the global state machine. Instead, all information is derived from the component state machines, which avoids a potential state explosion and lengthy test cases. Furthermore, the test suite checks for composition faults only. This substantially reduces the size of the test suite and thus the overall test effort.

- Post-processing Polygonal Voxel Data from Numerical Simulation (2004)
- Many applications dealing with geometry acquisition and processing produce polygonal meshes that carry artifacts like discretization noise. While there are many approaches to remove the artifacts by smoothing or filtering the mesh, they are not tailored to any specific application subject to·certain restrictive objectives. We show how to incorporate smoothing schemes based on the general Laplacian approximation to satsify all those objectives at the same time for the results of flow simulation in the application field of car manufacturing. In the presented application setting the major restrictions come from the bounding volume of the flow simulation, the so-called installation space. In particular, clean mesh regions (without noise) should not be smoothed while at the same time the installation space must not be violated by the smoothing of the noisy mesh regions. Additionally, aliasing effects at the boundary between clean and noisy mesh regions must be prevented. To address the fact that the meshes come from flow simulation, the presented method is versatile enough to preserve their exact volume and to apply anisotropic filters using the flow information. Although the paper focuses on the results of a specific application, most of its findings can be transferred to different settings as well.

- Embedding a Chained Lin-Kernighan Algorithm into a Distributed Algorithm (2004)
- The Chained Lin-Kernighan algorithm (CLK) is one of the best heuristics to solve Traveling Salesman Problems (TSP). In this paper a distributed algorithm is proposed, were nodes in a network locally optimize TSP instances by using the CLK algorithm. Within an Evolutionary Algorithm (EA) network-based framework the resulting tours are modified and exchanged with neighboring nodes. We show that the distributed variant finds better tours compared to the original CLK given the same amount of computation time. For instance fl3795, the original CLK got stuck in local optima in each of 10 runs, whereas the distributed algorithm found optimal tours in each run requiring less than 10 CPU minutes per node on average in an 8 node setup. For instance sw24978, the distributed algorithm had an average solution quality of 0.050% above the optimum, compared to CLK's average solution of 0.119% above the optimum given the same total CPU time (104 seconds). Considering the best tours of both variants for this instance, the distributed algorithm is 0.033% above the optimum and the CLK algorithm 0.099%.