The SPD experiment at the NICA collider involves not only the processing of multiple petabytes of data per year obtained from the detector, but also the production of similar amounts of data as part of the modeling of physical processes and expected signals from the front-end electronics. Because of this, the SPD experiment relies heavily on distributed computing for offline data storage and processing. This article provides the preliminary steps to build a distributed computing infrastructure for the SPD experiment, including the network backbone, storage and computing facilities at participating parties, their software components and configuration along with some higher-level software components necessary for the smooth operation of such infrastructure.
This is an English-language abstract. The article is devoted to evaluating the effectiveness of existing systems for identifying AI-generated text in scientific papers. The study analyzed six systems for their ability to detect AI-generated text in academic works. The probability scores for AI generation varied from 5 to 67
In this work, we propose and investigate multifactor strategies for managing the execution and scheduling of knowledge-intensive applications such as CyberShake, Epigenomics, SIPHT, Montage, LIGO, on cloud platforms “Workflow as a Service” (WaaS). A comprehensive solution to the problem of running applications in the WaaS model is considered using various criteria and optimization scenarios based on a cyclic scheme. The proposed scheme allows supporting a set of technologies and management tools for the execution of interdisciplinary knowledge-intensive applications on WaaS platforms.
In a number of works, polaron modeling is carried out with parameter values that accelerate the dynamics of the system. Earlier, analytical relationships of polaron characteristics were derived for the continuum model. The results of direct modeling in discrete chains with large-radius polaron parameters show good agreement with the analytics. Thus, using the obtained model values, it is possible to estimate the mobility and effective mass of the polaron for a wide range of parameters, including those corresponding to polyT DNA fragments.
Federated data analysis is a technology for building distributed data analysis systems where no data is moved from their storage (collection) locations for analysis. This is a data analysis technology that defines a new level of access and transfers analysis and calculations directly to where the data is located. Federated data analysis allows researchers to safely analyze data from different organizations. Restrictions on data access, cybersecurity requirements, and the development of edge device intelligence are all factors that will ensure growing interest in this technology. The paper examines issues of designing the architecture of such systems.
This paper describes a concept for a system designed to significantly simplify the development of distributed applications. From the perspective of a developer using this system, creating a distributed application is reduced to developing a single-instance application running on one computer. The system enables the developer to perceive the distributed infrastructure of the future application as a single computer with persistent memory, whose resources can be scaled up or down during the application’s operation.
An adaptive automated response system for Virtual Computer Lab and LMS Moodle is presented, leveraging Retrieval-Augmented Generation (RAG), a fine-tuned Llama (or Gemma, Qwen, etc.) model, and serverless architecture. Integrated with Moodle and Supabase, it delivers context-aware responses tailored to user roles (student, instructor, administrator). A self-learning mechanism driven by feedback enhances response accuracy and reducing technical support workload. An interactive interface with custom widgets improves user experience.
The architecture of an educational cloud environment designed for practical training of students in the field of DevOps and CI/CD is presented. The platform is based on open technologies: Gitea, Jenkins, Nexus, Kubernetes and Ansible. The goal of the project is to create a reproducible environment that mimics the full software development cycle. The possibilities of integration into the educational process and advantages over traditional approaches are shown.
The Jiangmen Underground Neutrino Observatory (JUNO) is a multipurpose neutrino experiment that will soon begin data taking. Its large data volume and international collaboration require a distributed computing infrastructure (DCI) across several sites. To ensure stable operations, we developed a new monitoring system that combines log collection, active service probing, workflow-based task management, visualization, and automated alerts. The system has been applied successfully in JUNO Data Challenges and is prepared for integration into routine operations. Future developments will focus on extending metrics, coupling with production systems, and expanding to other experiments.
The scientific paper presents a distributed data storage system deployed at the Joint Institute for Nuclear Research (JINR) to support participation in the neutrino experiment (JUNO). The storage architecture is based on the dCache middleware. Storage management is implemented using the GitOps methodology: declarative configuration descriptions, storage in Git, version control, and regulated deployment via GitLab CI/CD. Automated deployment and node classification are performed using Puppet and Foreman. The monitoring stack is built on Prometheus, Grafana, Loki/Alloy, and Alertmanager. The results of the system’s current operation are provided.
Managing a complex Linux-based production environment of the computing center is a highly challenging operational task. Such tasks require a high level of automation in distributed multi-component systems, which must be applied to complex operational workflows. There are several approaches to achieve this goal, including: creating operational scripts based on Linux shell commands and programming languages, using specialized software for specific tasks (backups, configuration, management) or employing advanced orchestration tools. Among these solutions, the StackStorm automation engine stands out. This paper describes the use of this platform for orchestrating operational workflows such as distributed backups, distributed system upgrades and distributed system administration in the Linux-based computing center at NRC “Kurchatov Institute”—IHEP.
The improvement of the server load balancer in distributed computing systems based on the BOINC platform is being considered. Such a grid system includes an application server, a database, a task scheduler, and client software. The efficient operation of the scheduler determines resource utilization levels and minimizes node downtime. The ComBoS simulator is used to analyze and refine scheduling algorithms by modeling the system’s behavior in a controlled environment. This approach helps avoid the costs and limitations associated with operating a real infrastructure. The experiments are based on historical data from the RakeSearch project, with variations in client computing power and task complexity. Analysis of the results allowed for minimizing task waiting times and increasing the efficiency of available computing power usage.
The paper proposes an integrated approach to configuring low-level server software for resource-intensive tasks of processing physical experiment data, in order to achieve a balance between maximum performance and minimum energy consumption. To implement the approach, a monitoring system that collects energy consumption and temperature metrics of key server components in real time has been implemented. A specialized analysis tool that allows you to correlate monitoring data with the results of load testing has been created. Calculated performance and energy efficiency coefficients for various settings are presented.
Due to the significant pace of development of hybrid supercomputers, universities face the task of training specialists to work with high-performance technologies. In 2016, a hybrid computing cluster was put into operation at Voronezh State University, the architecture of which provides extensive opportunities for teaching students parallel programming skills using OpenMP, MPI and CUDA technologies. The hardware and software features of the cluster are considered and strategies are proposed for the most complete implementation of available resources for training highly qualified personnel.
The task of determining the coastline from satellite images is reduced to the task of segmentation. DeepResUnet and TransUnet deep neural network models were selected. Landsat8 with images of rivers and bays in Russia was chosen as the dataset. Computational experiments were conducted to train two models on one node (deep learning), on multiple nodes (distributed deep learning and federated learning). The results of solving the problem of determining the coastline both on a single node and on a distributed system are presented. For experiments on federated learning, various methods of data separation between nodes of a distributed system have been identified. Data separation between nodes is proposed, both according to the geographical location of the satellite image and randomly. A comparative analysis of the results of federated learning with various methods of data distribution is discussed.
The paper discusses modifications to the load balancing subsystem in a distributed system based on the BOINC platform. It is proposed to use machine learning methods to estimate the complexity of subtasks and the probability of calculating a specific subtask at a specific node. The results of using a modified load balancing subsystem are presented using the example of solving the problem of constructing a spectrum of Latin squares of the order of 12 for the number of transversals.
This paper proposes practical methods for improving performance when using an evolutionary algorithm in a volunteer computing environment. The analysis showed that the low stability of some nodes leads to a delay in the creation of a new generation. The paper will propose a way to overcome this problem by maintaining a stable resource.The problem of the inability to pre-estimate task durations is also discussed and methods for overcoming it are proposed.
Computer administration of complex computing systems and distributed computing clusters presents significant challenges. Each sophisticated computer system has unique architecture and software configurations. Long-term system administration of such environments includes daily execution of various server commands - both for troubleshooting and for monitoring system behavior under different operational loads. These accumulated commands can be transformed into a valuable knowledge base for future use. This paper describes the development of such knowledge management system using Linux history tools at NRC “Kurchatov Institute”—IHEP.
The article will address the topic of encryption and backup of system data on the Heterogeneous HybriLIT platform [1].
