Abstract
Recent studies show that many real-world software faults are due to slight modifications (mutations) to the program. Thus, analyzing transformations made by a developer and associating them with well-known mutation operators can help pinpoint and repair the root cause of failures. This paper proposes a mutation operator inference technique: given the original program and one of its subsequent forms, it infers which mutation operators would transform the original and produce such a version. Moreover, we implemented this technique as a tool called Morpheus, which analyzes faulty Java programs. We have also validated both the technique and tool by analyzing a repository with 1753 modifications for 20 different programs, successfully inferring mutation operators 78% of times. Furthermore, we also show that several program versions result from not just a single mutation operator but multiple ones. In the end, we resort to real-world case studies to demonstrate the advantages of this approach regarding program repair.

Abstract

Abstract
Vulnerability detection and repair is a demanding and expensive part of the software development process. As such, there has been an effort to develop new and better ways to automatically detect and repair vulnerabilities. DifFuzz is a state-of-the-art tool for automatic detection of timing side-channel vulnerabilities, a type of vulnerability that is particularly difficult to detect and correct. Despite recent progress made with tools such as DifFuzz, work on tools capable of automatically repairing timing side-channel vulnerabilities is scarce. In this paper, we propose DifFuzzAR, a new tool for automatic repair of timing side-channel vulnerabilities in Java code. The tool works in conjunction with DifFuzz and it is able to repair 56% of the vulnerabilities identified in DifFuzz's dataset. The results show that the tool can indeed automatically correct timing side-channel vulnerabilities, being more effective with those that are controlflow based.

Abstract
The main goal of this systematic review is to synthesize existing evidence on the use of immersive virtual reality (IVR) to train professionals as well as to identify the main gaps and challenges that still remain and need to be addressed by future research. Following a comprehensive search, 66 documents were identified, assessed for relevance, and analysed. The main areas of application of IVR-based training were identified. Moreover, we identified the stimuli provided, the hardware used and information regarding training evaluation. The results showed that the areas in which a greater number of works were published were those related to healthcare and elementary occupations. In hardware, the most commonly used equipment was head mounted displays (HMDs), headphones included in the HMDs and handheld controllers. Moreover, the results indicated that IVR training systems are often evaluated manually, the most common metric being questionnaires applied before and after the experiment, and that IVR training systems have a positive effect in training professionals. We conclude that the literature is insufficient for determining the effect of IVR in the training of professionals. Although some works indicated promising results, there are still relevant themes that must be explored and limitations to overcome before virtual training replaces real-world training.

Abstract
Deep Learning (DL) training requires efficient access to large collections of data, leading DL frameworks to implement individual I/O optimizations to take full advantage of storage performance. However, these optimizations are intrinsic to each framework, limiting their applicability and portability across DL solutions, while making them inefficient for scenarios where multiple applications compete for shared storage resources. We argue that storage optimizations should be decoupled from DL frameworks and moved to a dedicated storage layer. To achieve this, we propose a new Software-Defined Storage architecture for accelerating DL training performance. The data plane implements self-contained, generally applicable I/O optimizations, while the control plane dynamically adapts them to cope with workload variations and multi-tenant environments. We validate the applicability and portability of our approach by developing and integrating an early prototype with the TensorFlow and PyTorch frameworks. Results show that our I/O optimizations significantly reduce DL training time by up to 54% and 63% for TensorFlow and PyTorch baseline configurations, while providing similar performance benefits to framework-intrinsic I/O mechanisms provided by TensorFlow.

Abstract
Distributed data management systems have increasingly been using variants of Snapshot Isolation (SI) as their transactional isolation criteria as it combines strong ACID guarantees with non-blocking reads and scalability. However, most existing proposals are limited by the performance of update propagation and stability detection, in particular, when execution and storage are disaggregated. In this paper, we propose TOPSI, an approach providing a restricted form of Parallel Snapshot Isolation (PSI) that allows partially ordering recent transactions to avoid waiting for remote updates or using a stale snapshot. Moreover, it has the interesting property of making a prefix of history in all sites converge to a common total order. This allows versions to be represented by a single scalar timestamp for certification and storage in a shared store. We demonstrate the impact on throughput and abort rate with a proof-of-concept implementation and the industry-standard TPC-C benchmark. © 2021 ACM.