Software & Systems

Latest issue articles

Order result by: Public date | Title | Authors

1. Increasing the efficiency of the brute-force algorithm for load distribution in hierarchies [№4 за год]
Authors: Aye Min Thike, Lupin, S.A., P.N. Telegin, Shabanov, B.M.
Visitors: 967
Finding the optimal distribution of tasks among the nodes of hierarchical systems is a complex combinatorial problem with many constraints. The efficiency of hierarchies in various areas of their application depends on the accuracy of its solution. This paper examines two approaches for constraining the solution search space, aimed at increasing the efficiency of the brute-force algorithm for load distribution in hierarchies. The proposed approaches are based on restructuring the process of generating a solution variant, aimed at excluding invalid combinations. In the first approach, during the generation of each new load distribution variant, enumeration for the final node is eliminated by replacing it with the computation of a single valid value. The second approach employs an early exit from the solution vector generation loop before its completion, triggered by the violation of any problem constraint. These methods are applicable either separately or jointly to improve the performance of the brute-force algorithm. The conducted computational experiments show that the first approach provides a 26-fold acceleration, the second approach provides almost a 3-fold acceleration, and their combination yields a 76-fold acceleration. These results confirm that the proposed approaches significantly reduce the computational complexity of the exhaustive search algorithm while maintaining the accuracy of the resulting solution. These approaches can also be implemented in parallel versions of the brute-force algorithm. This extends the algorithm's practical application scope to large-scale problems.

2. Supercomputer task runtime forecast using machine learning methods [№4 за год]
Authors: Barantsev, V.V., Mokryakov, A.V., Prilipko, A.A.
Visitors: 982
The paper applies machine learning methods to predict job execution times in supercomputer systems. The supercomputer job scheduler creates launch schedules based on user-provided runtime estimates. Users typically overestimate their jobs' execution time to avoid the risk of forced termination once the allocated time expires. This results in suboptimal schedule construction and significantly reduces overall scheduling efficiency. Job execution time prediction will enable the scheduler to generate more accurate schedules. The authors employed a comparative analysis of machine learning models as their research method, including decision trees, k-nearest neighbors, random forest, gradient boosting, neural networks, and broad learning. Model training utilized statistical data from job executions on the MVS-10P supercomputer. The study additionally examined approaches for improving prediction quality, including job clustering and classification methods. The research results revealed specific characteristics of applying machine learning for job execution time prediction with limited and often uninformative feature sets. The paper demonstrates that existing machine learning methods possess certain limitations related to model stability and overfitting risks. At the same time, the obtained results make it possible to identify potential ways to improve prediction accuracy. The practical significance of the study lies in the possibility of using its results to optimize job scheduling in supercomputing systems by increasing the accuracy of runtime forecasts.

3. Implementation of interoperability between heterogeneous quantum key distribution networks within an interuniversity quantum network [№4 за год]
Authors: Buzhin, I.G., Velikhov, V.E., Mironov, Yu.B., Ovsyannikov, A.P.
Visitors: 921
The paper focuses on developing and investigating a quantum key distribution (QKD) network model integrated into the national research and education network (NREN) infrastructure. The subject of the study is an interuniversity QKD network, implemented in 2024 on the basis of Russian NREN. The interuniversity QKD network includes QKD segments (networks) that use different protocols that are incompatible with one another, as well as interact with other QKD networks (the QKD backbone and university QKD networks). The paper explores the interaction of heterogeneous QKD networks (domains). The research investigates protocols for quantum-protected key transmission across multiple domains, encompassing inter-domain target key distribution through a core quantum network connecting distant locations. The authors proposed a practically oriented extended multi-level model of a quantum network. The model includes: the communication level, the synchronization level, the quantum key generation level, the quantum-secured key generation and management level, the QKD network management and monitoring level, the inter-domain interaction level, and the application level. The fundamental innovation lies in the inter-domain communication layer, responsible for universal addressing scheme implementation, protected distribution of routing intelligence and key material, and interoperability mediation be-tween cryptographic equipment and diverse quantum network implementations. A significant enhancement to the structural model is the dedicated synchronization layer. This enables a common frequency-reference framework for scalable QKD networks and establishes the foundation for a national time-and-frequency synchronization network for scientific research. The modification further enables higher admissible loss margins in quantum channels while preserving stable quantum key distillation performance. The paper also justifies a modular approach to the design of control and monitoring systems for scalable QKD networks.

4. Hypergraph-based modeling of complex technical systems for agent interaction determination [№4 за год]
Author: Zyablova, E.R.
Visitors: 923
The paper notes the expediency of using GH-graphs and GH-hypergraphs for modeling complex technical systems. It provides a description of a software package for modeling interaction between objects in complex systems using the Python programming language. The software package includes modules for dynamic visualization of the graph model, computation of graph characteristics, implementation of proportional graph partitioning algorithms, and shortest path search. The author uses the JSON format to represent graph structures. This paper is a development of the author's previous works, where the GH-graph is defined as a fuzzy graph with different types of vertices and multiple edges of different types. The advantages of the GH-graph include its ability to represent heterogeneous relationships in complex systems and its reduction of computation time for various characteristics through the use of multi-dimensional connection vectors that unify diverse relationship types. The author proposes a modification of the GH-hypergraph, which consists in the integration of the GH-graph into the hypergraph. The modified GH-hypergraph has the advantages of a GH-graph and additionally allows the use of different types of hyperedges and multiple edges of different types between a vertex and a hyperedge. This makes it possible to represent objects (groups of objects) of different types, different relationships between an object and a group of system objects, and reduce the time needed to analyze the system. Using the example of a specified extended perimeter security system, the paper demonstrates modeling capabilities for object interactions based on GH-graphs and GH-hypergraphs. As a practical example, the solution to the problem of forming zones of influence for system objects is considered. The method comprises system modeling stages along with computation of metrics for the GH-model and/or its components. Experimental results demonstrate at least a 1.3-fold reduction in metric computation time for models with up to 1000 vertices, compared to models that only support heterogeneous vertex connections. Future development of the proposed method can be achieved by addressing classification and forecasting tasks through predictive models based on explainable graph or hypergraph neural networks.

5. Generation and analysis of communication protocols using NK-automata and their modifications [№4 за год]
Authors: Kol’chugina, E.A. , Stezhka, V.A.
Visitors: 845
This study focuses on methods and algorithms that enable intelligent devices to explore and learn communication protocols in distributed infocommunication systems and environments. The research adopts an evolutionary approach to constructing NK-automata and their modifications. The objective is to obtain an NK-automaton capable of reproducing TCP protocol packet sequences through its state transitions. These sequence must demonstrate correctness when interacting with publicly available TCP echo servers and remain suitable for data transmission throughout TCP communication sessions. Creating such an automaton signifies that the intelligent device has learned the TCP protocol. To achieve this goal, the authors developed a modification of the classical NK-automaton called TCP-NK-automaton, along with software to simulate its operation. Unlike other NK-automaton variants, the proposed modification models a passive TCP packet data structure rather than an active software architecture. Experiments with the developed software successfully produced automaton instances that gen-erated correct TCP protocol packet sequences, simulating TCP sessions and demonstrating successful learning. The ob-tained results enable device compatibility through protocol learning without neural networks, eliminate the need for brokers, and enable the investigation and reverse engineering of previously unknown protocols.

6. Simulation of control modes for onboard space flying robots in virtual environment systems [№4 за год]
Authors: Strashnov, E.V., Kononov, D.A.
Visitors: 724
This paper considers the task of onboard space flying virtual model robot control in a virtual environment. To solve this task, the authors propose methods based on command, semi-automatic, and supervisory control modes for the robot. These modes characterize the degree of human operator participation in the robot control process. In command mode, the operator affects the controls (buttons, switches or levers) to execute commands, each of which is responsible for the motion of corresponding robot link. The semi-automatic mode assumes the operator's participation only in end effector control of the robot's manipulator. In supervisory mode, the operator sends commands, in response to which the robot automatically performs a certain sequence of actions, while the human monitors the progress of their execution. The proposed control approaches are implemented for solving navigation tasks within virtual space, capturing handrails with the robot's manipulator inside the space station, and docking with the docking station. Implementation of robot control modes includes constructing the robot's motion path, solving inverse kinematics, providing feedback based on virtual sensor readings, and computing the magnetic force to hold the robot during docking with the docking station. The authors test the developed approaches in the VirSim virtual environment software complex, demonstrating control of a virtual space flying robot model using a physical joystick. The simulation results showed that the supervisory mode is the most suitable solution for the space industry, while the command and semi-automatic modes are applicable for solving non-standard tasks under full human control. The authors conducted an analysis of flying robot motion during their simulation in a virtual environment. The practical significance of the paper's results lies in their ability to form expert opinions on using robots and their control methods for various tasks inside space modules.

7. Assessing the effectiveness and quality of placement solutions for VLSI circuit components [№4 за год]
Authors: Danilchenko, V.I., V.V. Kureychik
Visitors: 844
The paper considers the problem of placing fragments of very large-scale integration (VLSI) circuit components on a plane, taking into account technological and physical constraints. The problem formulation involves forming a set of design metrics, such as wirelength, density, total area, intersection index and others. The paper also examines a metric based on the number of linear segments. It serves to shorten connections and organize their configuration, leading to improved power efficiency and reduced signal transmission losses. They created a hybrid evolutionary model for a genetic algorithm, combining local adaptation by J. B. Lamarck with stochastic mutations by H. de Vries. Furthermore, the authors implemented an algorithm based on a stem cell behavior model. They introduced modifications aimed at improving placement quality, specifically targeting the minimization of intersections and the optimization of placement density. The authors also developed a software package to perform simulation and computational experiments using test sets. The effectiveness of the implemented and modified algorithms and metrics is evaluated using statistical methods, including correlation analysis. The paper evaluates the effectiveness of different metrics and demonstrates their application examples in design scenarios across various datasets. Following the research findings, the authors formulated recommendations for integrating design metrics into the VLSI design process. This integration enhances both the quality of design solutions and the manufacturability of the design process. This paper will benefit specialists and researchers in the field of integrated circuit design, as well as anyone interested in a multi-criteria approach to optimizing design procedures.

8. Dynamic forecasting of optimal resource allocation based on the trajectory continuation principle [№4 за год]
Authors: Kureichik, V.V., Semenov N.A.
Visitors: 666
The resource allocation problem has been thoroughly studied and solved as a mathematical programming task. However, the dynamic approach remains under development. The paper attempts to construct an algorithm for dynamic resource allocation forecasting. The classical variational problem is reduced to an optimal resource control problem and solved using the trajectory continuation principle. The forecasting problem is linear in the control input, with its key characteristic being the simplicity of vertex identification. This task remains highly relevant due to the persistent need to solve linear programming problems. During their research, the authors employed an approach based on the LIFO principle, which significantly simplifies and accelerates the polyhedron facet traversal procedure. The scientific novelty of the method lies in the combination of a dynamic approach with the LIFO rule, which reduces computational costs and improves forecasting accuracy. The method demonstrates its effectiveness in dynamic environments by enabling real-time adaptation of control strategies in response to system state changes. The paper also presents a modular scheme of the optimization program, illustrating the algorithmic and modular implementation of the method, which helps to better understand the process and the sequence of steps. The authors propose a concept of optimal solution to the problem. The paper highlights the importance of an integrated approach to resource management and offers new perspectives for further research. This enables the development of a more flexible and adaptive resource management system capable of promptly responding to changing external conditions and requirements. Such progress may subsequently lead to more efficient resource allocation and better performance results in multiple application areas.

9. Defining the set of user account attributes for centralized administration of heterogeneous systems [№4 за год]
Author: Efimov A.Yu.
Visitors: 853
To improve information security efficiency and minimize resource requirements, centralized management of protection mechanisms is used in complex information systems. The paper focuses on addressing user account management challenges, specifically organization of account attributes within heterogeneous information systems. The relevance of the issue is confirmed by problems due to differences in the implementation of security mechanisms, particularly user account management systems, across information system components. The author examines existing problem-solving methods and high-lights the role of user account attribute sets in ensuring interoperability within heterogeneous information systems. A new effective methodology for account management introduces attribute similarity assessment across different operating system platforms and subsequent categorization into universal and platform-specific groups. The paper describes four key components: a user account attribute set model for user accounts in heterogeneous systems; a corresponding methodology for attribute specification; evaluation of the approach's advantages and limitations; and practical application requirements and methods. The author also identifies directions for future research. Implementing this approach will simplify centralized administration of information security systems and reduce required administrative resources without compromising protection effectiveness.

10. Automatic synthesis of intelligent controllers based on a self-organizing robust knowledge base algorithm [№4 за год]
Author: Ignatyev, V.V.
Visitors: 910
This paper presents a newly developed program for the automatic synthesis of intelligent controllers based on a self-organization algorithm for robust knowledge bases. It also describes the implementation principles of the intelligent controllers under consideration, which are designed to achieve effective control of technical objects. The target control objects are systems describable by linear or nonlinear mathematical models of the first, second, or third order, including those with time delays, operating under uncertainty conditions. The research considers several uncertainty types: variations in control object parameters, external disturbances, and linguistic uncertainty. The program accounts for all these types. The system is developed using a novel mathematical framework, expressed through corresponding methods and algorithms implemented in the intelligent controller. The system delivers desired quality of control actions for objects operating under conditions of uncertainty. The program's foundation consists of scientific solutions for hybrid design of the intelligent controller's rule base, where the classical controller serves as the knowledge source for the fuzzy one. Using the acquired knowledge, the system synthesizes a fuzzy controller and trains it through neural networks and genetic algorithms. The resulting fuzzy inference system enables effective control of the considered classes of objects. The author demonstrates the performance of the developed program on an unstable nonlinear technical object of third order. The simulation results indicate the program's potential for designing intelligent controller rule bases that are synthesized automatically according to their structure. The source code, developed in the MATLAB programming language, is fully compatible with the author's preferred fuzzy logic design tools for intelligent control systems within this platform.

| 1 | 2 | Next → ►