Klaus Echtle

A Genetic Algorithm for Fault-Tolerant System Design

Due to high cost, considerable complexity and long design cycles of fault-tolerant systems, a (partial) automation of the design process becomes attractive. This paper presents an approach to automatic design by use of a genetic algorithm. Unlike typical genetic algorithms the individuals (which represent a fault-tolerant system structure each) are represented by a non-cyclic graph rather than a string. Special crossover and mutation operations modify the individuals such that reasonable fault-tolerant systems are likely to be generated. The biggest problem in using genetic algorithms lies in the definition of an appropriate fitness function one has to apply to each of the many generated individuals. A complete analysis of a single fault-tolerant system would comprise time-consuming fault-tree analysis, reachability analysis of the state space, etc. A substantial speed-up by orders of magnitude has been achieved by the development of a completely new fitness function, which can be considered as a simplified reachability analysis. For many fault tolerance techniques it visits each component only once (or very few times in the case of mechanisms like rollback, retry etc.).

SAL-Based Symbolic Scheduling in Time-Triggered Networks

This paper presents a novel approach for combined task and message scheduling for TDM-based avionics applications that allows to automatically compute schedules with minimal end-to-end latency. Our approach relies on a symbolic encoding of the scheduling problem. The symbolic encoding is done by constructing gradually a schedule beginning in a state where no task has started and no messages have been sent on the bus yet, and proceeding one step at a time assigning starting times to tasks and slot positions to messages. We use the SAL toolset from SRI for our experiments. Experimental results demonstrate how the latest generation of model-checking tools meet the challenges of providing both a convenient modeling language and the performance to solve given scheduling problems.

Clock synchronization issues in multi-cluster time-triggered networks

We address the issue of establishing and maintaining a system-wide common time base in fault-tolerant multi-cluster time-triggered systems. We propose an approach how to synchronize system nodes among several clusters using the fault-tolerant mid-point algorithm. Before executing clock synchronization each node measures the clock deviation values and stores them in a convenient data structure. From these values the clock synchronization algorithm calculates a correction term which should be added or subtracted from the local clock. For distributed real-time systems that are structured in a set of clusters the set of clock deviations can be subdivided into a set of local clock deviations and a set of global clock deviations. Local clock deviation values (respectively global clock deviation values) of a specific node are captured by building the time difference between the observed and expected arrival time of synchronization messages sent by a node belonging to the same cluster (respectively to another cluster). In order to receive messages from other clusters the clock deviation between the sender and the receivers should be bounded. We derive the lower bound of the network precision of a multi-cluster system that executes the FlexRay protocol and will show that it depends mainly on the transmission delays and measurement errors. Further, we inquire about the amount of the minimum time gap between two successive messages that could be exchanged via the FlexRay System. This time gap is an important parameter for developing a correct configuration of multi-cluster systems.

Avoiding Malicious Byzantine Faults by a New Signature Generation Technique

Agreement problems like interactive consistency, reliable broadcast, group membership, etc. require a high protocol overhead when they must be solved under general (and thus hard) fault assumptions. Known signature methods contribute to more efficient solutions by protecting forwarded information from being altered undetectably. This paper presents a new signature generation technique, which prevents the occurrence of malicious Byzantine faults in the sender with very high probability. Hence, it is not necessary to exchange multicast messages among the receivers for an equality check. This advantage opens an extended design space of agreement protocols with fewer messages, fewer timeouts and thus lower execution times. The new unique signature generation algorithm (called UniSig) is based on alternately stepwise generation of coded sequence numbers and digital signatures. Different messages cannot obtain the same valid signature, because the steps to increment the coded sequence number are included in UniSig. Deviations from the program execution path are very likely to lead to detectably corrupted signatures. Hence, for each sequence number a valid signature can be generated only once.

Simulation-based reliability evaluation for analog applications

In this paper we present a new technology to improve the reliability evaluation for analog circuits based on fault simulation. A fault simulation framework has been developed and applied to reliability prediction in practice. In contrast to traditional reliability prediction the evaluation is not based on the “serial system approach”, which is mostly too pessimistic. Instead, the new approach considers all types of redundant structures within the system, even hidden redundancy and fault tolerance. The reliability prediction is more accurate because it is based on fault simulation results and the quantitative assessment of tolerated faults. This also results in a detailed reliability model related to practical reliability prediction in early design stages. By EDA integration the practical relevance has been highlighted.

Simulation of Clock Synchronization in Multi-Cluster FlexRay Systems

This paper deals with the simulation of Clock Synchronization in Multi-Cluster systems that use FlexRay, a communication protocol designed by the FlexRay industry consortium in order to meet the fault-tolerance and dependability requirements of distributed real-time applications. Using SiMuFlex (Simulation model for Multi-Cluster FlexRay Systems), we investigate the performance of the clock synchronization algorithm and determine the amount of the system precision which is a measure for the quality of the clock synchronization. The simulation experiments are focused on the evaluation of the convergence and stability of the clock synchronization algorithm when the clock drift rates of several nodes undergo systematic variations. Moreover, we investigate the impact of the blackout and delays on the gateway component which interconnects the system clusters.

Automatic design of reliable systems consisting of nano-elements

The design of nano structures is considered a challenge for design methods, which have to cope with much more elements than traditional VLSI. Moreover, such elements will be unreliable due to unavoidable physical quantum effects. The inevitable fault tolerance leads to an additional increase of the design space. An extremely high number of nano-devices can be used for various redundancy schemes and many combinations thereof, thus promising an efficient solution to the reliability problem of nano devices. This paper proposes a heuristic design method based on a specific type of genetic algorithms. It has been adapted to the design of fault-tolerant nano systems with respect to the representation of systems as well as the fitness function and the underlying fault model.