Machine learning (ML) has become a crucial component in safety-critical systems, such as those used in autonomous vehicle perception. However, the correctness and, therefore, the safety of these systems can be compromised by out-of-distribution data, accidental faults, and security breaches. This paper investigates using a replicated ML architecture to mitigate the risks associated with complex single-points-of-failure. Additionally, it explores the application of rejuvenation to sustain healthy majorities when facing persistent threats. We evaluate the output reliability of the proposed architecture in two case studies: traffic sign detection and perception for autonomous driving. We adopt models and reliability functions, validating our findings using realistic data sets and fault injection experiments. We also evaluate driving safety using the proposed architecture in the CARLA simulator. Our results show that our models can present a good generalization and multi-version ML with proactive rejuvenation can improve correctness and, thus, safety despite faults and cyberattacks.

ABSTRACT Objective: NoSQL database management systems (DBMSs) are designed to handle large‐scale data for modern applications. These systems often operate in a distributed manner, allowing data to be spread across multiple nodes to ensure replication, reduce data loss, and facilitate recovery. The consistency level in these DBMSs determines how synchronized data is across nodes, influencing the trade‐off among consistency, availability, and system performance. Given that different applications have unique requirements, understanding the impact of various consistency levels is essential. This study conducts an in‐depth analysis of how consistency level choices affect the performance of three leading NoSQL DBMSs: Cassandra, MongoDB, and Redis. Methods: These systems were evaluated under different consistency configurations, user loads, and workloads, with performance metrics including average response time and operations per second. Results: Our results show that Cassandra and Redis handle write operations faster than MongoDB, though Cassandra experiences significant slowdowns of up to 200% when switching to strong consistency. This performance degradation is observed for both read and write operations, making Cassandra the most affected DBMS when opting for strong consistency. Conclusion: The detailed findings offer valuable insights into the trade‐offs between performance and consistency in these DBMSs, providing guidance for database engineers in selecting appropriate consistency levels based on their application needs.

Cooperation among autonomous vehicles is required whenever efficiency or safety prevents maneuvering based solely on the information of individuals. Intersection crossing is a prominent example of such a situation, where obstructed views create safety concerns and where driving on sight would lead to known inefficient solutions. However, communication, a prerequisite for cooperation, and, in general, the complexity of autonomous driving stacks elevate the threat surface beyond justifiable thresholds, creating the potential for cyberattacks to succeed, particularly when targeting the “brain”. Some of these attacks go undetected and may harm passengers, pedestrians, and other traffic participants in a vehicle's proximity. In this paper, we address a fundamental challenge of intrusion-resilient maneuver planning: the question of forming consensus groups given variations in the number$N$of vehicles that participate in complex maneuvers and given that in a larger group of cars, a larger number$F$may have already been compromised by an adversary. Introducing confirmed-location-based group membership, we show how trust-anchor-provided precise location information can be leveraged to establish a ground truth about N and$F$to efficiently solve and agree upon intersection crossing as representative of other complex maneuvers in an$F$fault-and-intrusion tolerant manner.

The use of traditional defense mechanisms or intrusion detection systems presents a disadvantage for defenders against attackers since these mechanisms are essentially reactive. Moving target defense (MTD) has emerged as a proactive defense mechanism to reduce this disadvantage by randomly and continuously changing the attack surface of a system to confuse attackers. Although significant progress has been made recently in analyzing the security effectiveness of MTD mechanisms, critical gaps still exist, especially in maximizing security levels and estimating network reconfiguration speed for given attack power. In this paper, we propose a set of Petri Net models and use them to perform a comprehensive evaluation regarding key security metrics of Software-Defined Network (SDNs) based systems adopting a time-based MTD mechanism. We evaluate two use-case scenarios considering two different types of attacks to demonstrate the feasibility and applicability of our models. Our analyses showed that a time-based MTD mechanism could reduce the attackers' speed by at least 78% compared to a system without MTD. Also, in the best-case scenario, it can reduce the attack success probability by about ten times.

Modern networks are dynamic with configuration changes that introduces a set of challenge to the network administrator in terms of security and availability. Here, the major challenge faced by the administrator is the increasing number of vulnerabilities with the uncertainties related to defense deployment options and how these options affect the network availability over time. This work proposes a hierarchical model-based approach to evaluate the availability of dynamic networks considering the deployment of different hardening options. In particular, this work adopts reliability block diagrams and Petri nets to represent and analyze dynamic network environments and evaluate their availability. A case study is presented to demonstrate the feasibility, usefulness, and scalability of the proposed approach for computing the availability of dynamic networks considering different hardening options. The proposed approach can be helpful for network administrators who are in charge of choosing the best hardening options taking into account the impacts on availability.

Systems unavailability may produce severe consequences for modern business such as data loss, customer dissatisfaction, and subsequent revenue loss. Disaster recovery (DR) solutions have been adopted by many organizations as an attempt to prevent data loss and ensure business continuity. With the cloud computing expansion, different cloud providers have been offering low-cost solutions for DR purposes such as the Backup-as-a-service (BaaS) for consumers. Therefore, in this paper, we present an integrated model-experiment approach to evaluate a BaaS environment for DR purposes. We use analytic models and fault-injection experiments to evaluate DR keymetrics such as availability, downtime, Recovery Time Objective (RTO), and Recovery Point Objective (RPO) in a real-world BaaS environment. The results revealed that the environment availability can vary according to the amount of data to backed up and restored. Besides, a sensitivity analysis shows that the RTO and RPO are mainly influenced by the the mean time to recover from a disaster and the backup interval, respectively.

Relational databases are the most popular database system worldwide. The occurrence of failures in these systems may produce severe consequences for the business, such as data loss, customer dissatisfaction, and subsequent revenue loss. Consequently, many organizations have adopted disaster recovery (DR) solutions as an attempt to prevent data loss and ensure business continuity. Data replication for databases is one of the most used DR solution employed to guarantee data safety and availability. However, the analysis regarding DR aspects has been less explored. Therefore, in this paper, we present an integrated model-experiment approach to evaluate replication mechanisms in relational databases for DR purposes. We performed experiments in a geo-distributed cloud environment and developed analytic models to evaluate DR key-metrics such as availability, downtime, Recovery Time Objective (RTO), and Recovery Point Objective (RPO). The results revealed that the adoption of replication mechanisms could increase the system's availability significantly. It also revealed that the replication mechanisms can guarantee RPO and RTO within seconds.

The Information Technology (IT) systems of most organizations must support their operations 24 hours a day, 7 days a week. Systems unavailability may have serious consequences such as data loss, customer dissatisfaction, and subsequent revenue loss. With the popularity of cloud computing, the adoption of cloud-based disaster recovery (DR) solutions has gained more space to prevent data loss and ensure business continuity. However, disaster recovery solutions are not cheap and do not exist as a single solution that suits every requirement (e.g., availability and costs). In this paper, we present an integrated model-experiment approach to evaluate cloud-based disaster recovery solutions. We use Stochastic Petri Nets (SPNs) and fault-injection experiments to evaluate availability related metrics like steady-state availability and downtime. To demonstrate the feasibility of our approach, distinct real-world cloud-based DR solutions (e.g., active/active and active/standby) are modeled and analyzed. The results revealed that disaster recovery solution significantly improves system availability and minimizes the downtime costs. In addition, our numerical analysis shows the statistical correspondence between the results of the experiments and models.

The demand for mobile devices and applications through the marketplaces has increased dramatically in recent years. Mobile applications take on even greater prominence and importance in the IT industry. Along with the growth of mobile technology, battery lifetime has become one of the most relevant challenges for this kind of device in the last few years. This paper presents a study assessing performance and energy consumption for mobile applications by means of stochastic models. In order to assist inexperienced system designers to create formal models, mapping rules were adopted to transform SysML diagrams into Deterministic and Stochastic Petri nets. The results of this work can help system designers in the design-making process in order to develop more efficient mobile applications. The estimates obtained from the models show that the proposed approach is indeed a good approximation to the respective measures obtained from the real infrastructure.