High Performance Computing (HPC) defines a computing paradigm characterized by extraordinarily powerful computing capacity. An HPC infrastructure is also equipped with high-bandwidth network connections, and extensive storage. The mentioned resources make their employment in academic and industrial environments clear, as well as the manipulation of huge amounts of sensitive and critical data. Since HPC infrastructures involve a broad quantity of desirable resources, including confidential data, and are employed for research and industry tasks, they become an attractive target for malicious entities. Furthermore, the heterogeneous resources involved in the HPC systems and their consideration as “trusted” systems worsen the situation. Applying traditional security measures to individual nodes of the infrastructure is an inadequate strategy because it does not consider the system as a whole. To enhance security, use a “defense-in-depth” strategy by implementing multiple layers of controls across the system, ensuring that if one or two layers fail, others remain in place to thwart potential attacks. While not ensuring absolute security, it makes successful attacks highly challenging. Clearly, this approach does not provide absolute security but only offers the means to make a successful attack so difficult to execute that it becomes unsustainable for an attacker. These considerations push the necessity to identify the threats that an HPC infrastructure may be affected to and how the security of these systems may be assessed to understand: the security level of the analyzed infrastructure; the applied security controls, if they are implemented; and the possible ones that may be applied to improve the security level. By following this direction, with this work, we propose: (i) a threat analysis for an HPC system through a fine-grained threat modeling technique; and (ii) a possible approach to perform a security assessment of an HPC system. Additionally, we propose a possible approach to selecting suitable security controls to improve the security level of an HPC infrastructure. Lastly, we present some results for V:HPCCRI: the supercomputer of the University of Campania Luigi Vanvitelli.
Mobile Edge Clouds (MECs) are emerging as a key complement to centralized cloud infrastructures by bringing computing and storage resources closer to the network edge, thereby reducing network bandwidth, latency, and jitter. A critical challenge in leveraging MECs effectively is the application placement problem, which seeks to minimize operational costs while ensuring end-user Quality of Service (QoS). This problem is further complicated by user mobility, as applications must migrate to maintain optimal QoS, yet frequent migrations can lead to unnecessary bandwidth consumption due to state transfer. In this paper, we tackle the application placement problem for stateful applications in MEC environments. We model the dynamic workloads, applications, and infrastructure typical of MECs and define the associated costs: resource utilization, migration, and QoS degradation. Based on this model, we propose two online placement algorithms – Gale-Shapley-based and Follow-me – designed to minimize the total cost of operating applications. These algorithms are compared against an offline benchmark that has complete future knowledge. Experimental results demonstrate that both proposed algorithms efficiently place applications in MECs, achieving operating costs within 8
Rapid evolution characterizes modern software systems, particularly evident with adopting continuous integration and delivery processes. However, while tools for maintaining monolithic architectures are well-established, there is a notable deficiency in methodologies for analyzing and managing changes within decentralized, microservice-based systems. As microservices increasingly become the backbone of cloud-native enterprise solutions, understanding the intricacies of how changes affect these systems becomes crucial. This paper investigates the impact of dependencies on the maintainability of microservice architectures and emphasizes the importance of managing these dependencies to prevent deterioration in system maintainability. We advocate for a systematic approach to dependency management that addresses the actual pathways through which changes propagate, providing a concrete alternative to traditional methods that often focus on symptomatic treatments such as anti-patterns and code smells.
Currently, energy consumption and energy efficiency are some of the concerns of cloud system administrators. In recent years standards such as ISO/IEC 30134-4 and ISO/IEC 21836 have emerged to improve energy efficiency. Both standards are focused on the evaluation of physical servers, considering the power consumed and the maximum peak of performance, under running a SPEC benchmark. Then, server consolidation through virtualization is not considered in these standards. However, the QoS is another point that is important to consider, at the same time, by data centers administrators. In this work, we proposed a standard methodology to quantify the trade-off between the energy efficiency and the QoS of consolidated servers. As a result, it has been demonstrated through real experimentation that the proposed methodology considers the consolidation of servers in any type of virtualization environment. Besides, this methodology helps system administrators to manage cloud data centers and servers more efficiently.
Background: Container orchestration systems such as Kubernetes rely heavily on declarative manifest files that serve as blueprints for orchestration. However, managing these manifest files often presents significant challenges and requires considerable expertise in DevOps. Methodology. This study explores the use of Large Language Models (LLMs) to automate the generation of Kubernetes manifest files using natural language specifications and prompt engineering techniques. We evaluate the effectiveness of these LLMs through Zero-Shot, Few-Shot, Prompt-Chaining, and Self-Refine methods to fulfill DevOps requirements and facilitate fully automated deployment pipelines. Results. The results indicate that LLMs can produce Kubernetes manifests with varying degrees of manual input, with GPT-4 and GPT-3.5 demonstrating potential for fully automated deployments. Interestingly, smaller models sometimes outperform larger ones, challenging the assumption that larger models are always superior. Conclusion: The research highlights the critical role of prompt engineering in enhancing LLM outputs for Kubernetes and suggests further research into prompt strategies and LLM performance comparisons, presenting a promising direction for integrating LLMs into automated deployment workflows.
