Area and reliability optimization have become the primary constraints in circuits logic synthesis. To address the increasing area and transient fault susceptibility in combinational circuits, we propose a high-dimensional genetic algorithm (HGA). HGA adopts an evolutionary scheme based on ternary tree, and uses adaptive crossover operator and flight operator to jump out of local optimum. Moreover, based on the HGA, we propose an area and reliability optimization method (AROM) for mixed polarity Reed-Muller logic circuits, which searches the best polarity with minimum area and soft error rate. The experimental results confirm that AROM can search for more desirable nondominated solutions in less time compared to existing optimization methods, and can be used as an effective electronic design automation tool for multi-objective optimization.

This paper presents a process-oriented GCC failure analysis method based on fault injection. We first analyze the process of GCC, systematically define the fault modes of GCC, and establish a GCC fault mode library. We also propose a software fault injection method based on SystemTap and develop a related fault injection tool. Subsequently, we proceed to carry out a series of fault injection experiments on GCC, aiming to unearth potential reliability issues. These experiments serve as a means to identify modules within GCC that are susceptible to failures and to propose potential enhancements. The results of these experiments unequivocally underline the existing problems in fault tolerance mechanisms across various aspects of GCC. These findings underscores the significance of our approach in failure analysis and reliability evaluation of GCC. Through a holistic analysis of diverse fault modes and their associated manifestations, our approach contributes to a more profound understanding of GCC failure behavior. It offers valuable insights to guide future optimization endeavors.

The high overhead of tracing, the amount of up-front effort required to select trace points, and the lack of effective data analysis model are the significant barriers to the adoption of intra-component tracing for fault diagnosis today. This paper introduces a novel method for fault diagnosis by combining function level adaptive tracing, fault injection, and graph convolutional network. In order to implement this method, we introduce techniques for (i) selecting function level trace points, (ii) constructing approximate function call trees for programs when using adaptive tracing, and (iii) constructing graph convolutional network with fault injection campaign. We evaluate our method on four widely used open source server software: Redis, Nginx, Httpd, and SQlite. The experimental results show that our method outperforms log-based method, full tracing method, and Gaussian influence method in terms of accuracy, efficiency, and performance impact on the diagnosis target.

Software fault injection is a fundamental technique for studying the dependability of software systems. The code mutation based defect injections encounter efficiency (i.e., dormant faults) and accuracy (when they are not done at the source code level) issues. Software error injections directly emulate the effects of software defects, which is more efficient and practical. But, whether the injected errors truly represent the impact of bugs in the real world is still an open problem. Several studies have investigated the representativeness problem of component interface (API) errors. However, the component API error injections are relatively coarse in granularity because they regard the component programs as black boxes. Thus, they cannot be used to analyze the internal error behaviors within the components. To overcome this issue, we propose a finite state machine for generating representative code mutations (as baseline) and a novel tracing approach for collecting function call interface data. That enables the analysis of the representativeness of program internal errors from the perspective of function call interface for C/C++ programs by fault-free and fault injection control experiments. The experimental results imply that the traditional error models cannot accurately emulate software defects at function call interfaces. We provide useful suggestions based on our findings for improving the representativeness of function call interface error injection.

Anomaly detection plays an important role in large-scale distributed systems. System logs are significant source of troubleshooting and problem diagnosis. Most of the existing anomaly detection methods apply only one machine learning model to extract feature from structured logs. However, each machine learning model has different strength towards different target system. It is hard for developers to know which is the best method to their practical problem at hand. This paper proposes two methods for anomaly detection based on the machine learning ensemble models. The first method takes mixture of experts to combine the weighted prediction of ensemble members to generate the final prediction. The second method divides the sample into n parts, then use n models to extracting features. Experimental results demonstrate the validity and accuracy of the proposed methods.