Volume 3(72)

CONTENTS

1. Rahmani J., Harisova D. R. Autotuning Web Application Performance by SLO: Observability, Diagnostics and Safe Optimizations
2. Sokolova O. D., Kratov S. V., Tkachev K. V. Environment Emergency Monitoring Systems: An Analytical Review of Publications   
3. Kulikov I. M. Technology for Solving the Riemann Problem Using the Harten - Lax - van Leer Method
4. Monakhov O. G., Monakhova E. A. Construction and Optimization of Topological Structures of Regular Networks with Parametric Description
5. Perepelkin V. A., Titov I. I., Podkolodny N. L., Matus N. E., Shevelev E. I., Kolchanov N. A., Marchenko M. A. Active Knowledge Base Construction for Automatic Analysis of Secondary RNA Structure

J. Rahmani, D. R. Harisova

Moscow Technical University of Communications and Informatics,111024, Moscow, Russia

AUTOTUNING WEB APPLICATION PERFORMANCE BY SLO: OBSERVABILITY, DIAGNOSTICS AND SAFE OPTIMIZATIONS

The article presents an approach to semi-automated performance improvement of web applications based on a rule-based pipeline that integrates service targets (SLO) and observability data (metrics, logs, traces). The methodology includes the identification of problem areas by symptoms, a catalog of optimizations and a procedure for experimental verification of the impact on the 95th percentile of delay, error rate and operating costs. The focus is on applications for energy companies. The novelty lies in the criteria for choosing optimizations that take into account not only acceleration, but also the risk of rollback, cost and impact on sustainability, as well as in the A/B testing protocol with bootstrap confidence intervals. Experimental testing on the scenarios of entering readings, payments, account access, and registration of requests showed a decrease in p95 latency by 48–55 % (bootstrap 95 % CI for Δp95 from 1.4 to 2.3 s), error rate by 70–93 %, and operating costs by 40–58 % (for individual operations) relative to baseline. The aggregated savings amounted to 42 %.

Key words: web applications, performance, service targets, observability, autotuning, bottleneck diagnostics, A/B testing, latency percentiles, secure optimizations, DevOps, SRE.

References

Bibliographic reference: Rahmani J., Harisova D. R. Autotuning Web Application Performance by SLO: Observability, Diagnostics and Safe Optimizations //"Problems of informatics",  2026, №3, pp..5-14.  DOI: 10.24412/2073-0667-2026-3-5-14.

O. D. Sokolova, S. V. Kratov, K. V. Tkachev

Institute of Computational Mathematics and Mathematical Geophysics SB RAS, 630090, Novosibirsk, Russia

ENVIRONMENT EMERGENCY MONITORING SYSTEMS: AN ANALYTICAL REVIEW OF PUBLICATIONS

This article presents the analytical review of recent scientific publications describing modern environmental emergency monitoring systems. The relevance of this research is driven by the global increase in natural and man-made disasters associated with climate change, intensified industrial production, and increasing anthropogenic pressure on ecosystems. Under these circumstances, timely and reliable information on the state of the atmosphere, hydrosphere, and forests is crucial for timely response and mitigation. Many studies (both Russian and international) describe a key shift in monitoring systems: a transition from fixed ground stations to intelligent distributed sensor networks based on the Internet of Things (IoT), cloud platforms, and Earth remote sensing (ERS) technologies, which enable the collection and processing of highly accurate data in real time.

The first section of the article analyzes publications describing the general principles of building modern environmental monitoring systems. The authors of these studies emphasize that the effectiveness of systems depends on the integration of three key components: sensor networks, remote sensing systems (including satellite constellations and unmanned aerial vehicles), and geographic information systems (GIS) for storing, analyzing, and visualizing spatial data. The analysis of publications in which the authors describe space-based monitoring systems reveals that multi-agent models play a significant role in system development, enabling the integration of information from various satellites and ground stations to improve management efficiency and optimize resources.

The second section presents a review of publications describing systems for detecting and predicting forest fires — one of the most common and dangerous natural disasters. The authors of these studies point to a shift from traditional satellite monitoring to intelligent platforms based on machine learning and convolutional neural networks. Innovative solutions are discussed, such as energy-efficient sensor networks based on the LoRa-Mesh protocol and the use of the Google Earth Engine cloud geospatial platform to calculate spectral indices of burned areas and vegetation.

The third section describes the problem of monitoring oil spills, which pose a threat to marine and freshwater ecosystems. The review of publications on modern oil spill monitoring systems is presented. Research on this topic describes ground-based and airborne monitoring methods that provide high- quality identification of hydrocarbons on various surfaces. The authors of all the articles on oil spill monitoring note a trend toward hybrid systems that combine satellite data with ground-based sensor networks and deep learning algorithms for automatic spill recognition and drift prediction. Various monitoring methods are classified by their physical principle and deployment platform.

The fourth section analyzes recent publications describing flood monitoring systems. Modern systems integrate hydrological sensors, meteorological radars, and satellite data. This review presents

articles describing two types of monitoring systems: traditional land-based hydrological stations providing real-time water level and discharge data, and modern Internet of Things based solutions using low-cost ultrasonic sensors integrated with geographic information systems. The use of remote sensing data for flood boundary mapping, particularly in cloudy and nighttime conditions (radar imaging), is analyzed. Recent publications describing methods for flash flood forecasting based on the numerical solution of shallow water equations and the use of recurrent neural networks for the analysis of time series of hydrological data are presented.

The fifth section summarizes the experience of applying artificial intelligence (AI) methods in environmental monitoring systems, as described in recent publications. Publications describing the use of convolutional neural networks for interpreting remote sensing data, climate scenario models, and intelligent decision support systems are analyzed. As noted by the authors of most studies, machine learning and deep learning algorithms significantly improve the accuracy of natural disaster forecasting, the automation of satellite image processing (segmentation, classification, object detection), and the identification of pollution sources in real time. Some papers point to existing challenges in applying AI methods in modern monitoring systems, including a shortage of experts, data access and privacy issues, and the need to develop hybrid models that combine neural network approaches with traditional mathematical methods.

At the end of the article, the main methods used in environmental monitoring systems are systematized in tables. In the Conclusion, the authors note that one of the key functions of modern systems is the collection and analysis of large data sets in real time to predict emergencies and identify natural and man-made hazards. The most promising areas for the development of environmental monitoring include the development of universal automated platforms, the implementation of energyefficient wireless networks, and the creation of hybrid predictive models.

Thus, creating an effective environmental monitoring system requires an integrated approach combining advanced sensor technologies, remote sensing tools, and artificial intelligence methods.

Key words: environmental monitoring, Earth remote sensing, geographic information systems (GIS), data collection and transmission networks, artificial intelligence (AI) methods.

This work was carried out under state contract with ICMMG SB RAS FWNM-2025-0005.

References

Bibliographic reference: Sokolova O. D., Kratov S. V., Tkachev K. V. Environment Emergency Monitoring Systems: An Analytical Review of Publications //"Problems of informatics",  2026, №3, pp.15-39.  DOI: 10.24412/2073-0667-2026-3-15-39.

I.M. Kulikov

Institute of Computational Mathematics and Mathematical Geophysics SB RAS, 630090, Novosibirsk, Russia

TECHNOLOGY FOR SOLVING THE RIEMANN PROBLEM USING THE HARTEN - LAX - VAN LEER METHOD

Using a Harten-Lax-van Leer-type method with a contact discontinuity for solving the Riemann problem within the Godunov scheme for hyperbolic equations is one of the most efficient approaches. This efficiency is achieved by specifically accounting for the tangential discontinuity in a continuous medium, where the error lacks a numerical mechanism for self-generation, leading to a significant reduction in the numerical dissipation of the overall Godunov scheme. This paper presents a unified technology for constructing the Riemann problem solution for the equations of ideal gas dynamics, ideal magnetohydrodynamics, special relativistic gas dynamics, and special relativistic magnetohydrodynamics, which covers most mathematical models of cosmic plasma evolution. The technology is based on formulating the equations as a symmetric hyperbolic system, for which a unified scheme to calculate the relations at the contact discontinuity is proposed, incorporating a piecewise parabolic reconstruction of physical variables. The verification of the developed mathematical apparatus was carried out on classical one-dimensional tests that admit analytical solutions.

Key words: Godunov Method, HLL/HLLC Solvers, Riemann Problem.

The research was carried out within the state assignment of the ICMMG SB RAS FWNM-2025-0005.

References

Bibliographic reference: Kulikov I. M. Technology for Solving the Riemann Problem Using the Harten - Lax – van Leer Method //"Problems of informatics",  2026, №3, pp..40-59.  DOI: 10.24412/2073-0667-2026-3-40-59.

O. G. Monakhov, E. A. Monakhova

Institute of Computational Mathematics and Mathematical Geophysics of SB RAS, 630090, Novosibirsk, Russia

CONSTRUCTION AND OPTIMIZATION OF TOPOLOGICAL STRUCTURES OF REGULAR NETWORKS WITH PARAMETRIC DESCRIPTION

Algorithms for constructing and optimizing promising classes of topological structures — parametrically defined regular graphs (R graphs) — are considered. A new model of communication network topologies for multiprocessor systems and networks-on-chip is proposed: a class of multi-level parametrically defined regular MR networks (MR graphs). Well-known regular graphs, previously proposed as computing system structures, can be used as level elements, combining them optimally. In this paper, parametrically defined regular graphs with a compact parametric description and periodic symmetry are considered as construction elements for generating multi-level networks. These graphs can be combined using the operation of multi-level graph composition.

This paper focuses on the construction and optimization of ordinary (simple) and multi-level R graphs. A recent example of the construction and optimization of multi-level networks is the application of the composition operation to circulant and chordal graphs [5, 6]. We extend this approach and show how similar ideas can be implemented within a more general class of R graphs and how to optimize the resulting simple and multi-level R networks using differential evolution on a compact parametric representation. We note that the compact parametric representation of R graphs presented in this paper allows us to reduce the complete description of the graphs by several orders of magnitude compared to a description using an edge list.

The goals of this paper are (1) to provide a simple parametric description of simple and multi-level networks based on R graphs, (2) to apply the multi-level composition operation to construct multilevel networks based on R graphs, (3) to obtain optimal simple and multi-level R networks for a given number of vertices N and degree v with a minimum average distance smaller than the average distance of optimal circulant networks of the same size, and (4) to compare R graphs by diameter and average distance with circulant graphs and multi-level circulant graphs.

In synthesizing simple and multi-level R networks, a differential evolution algorithm was used to determine the optimal parameters of the generated topology that minimize the average network distance for a given number of nodes, node degrees, number of symmetry classes, and number of levels. The optimal network synthesis algorithm was developed using large language models and implemented on a Kunpeng 920 processor.

The characteristics of the resulting simple and multi-level R networks were compared with known optimal circulants [3, 14] based on the following three metrics: (1) change in the average network distance with an increase in the number of nodes for degree 4, as the simplest and most studied class of circulants; (2) dependence of the average distance on the number of classes in the R network; (3)

Supported by the state assignment of ICMMG SB RAS N FWNM-2025-0005.

comparison of the average distance of R networks for the number of nodes from 25 to 210 and degrees from 4 to 10 with the optimal circulants obtained in [3]. The obtained results show a relative decrease in the average distance for optimal R networks compared to optimal circulants (in %), for the specified degrees and numbers of vertices. The graphs show that optimal R networks have a 13-24 % lower average distance than optimal circulants. Thus, it is demonstrated that the constructed R networks have better structural characteristics than circulant networks and multi-level circulant networks with the same equipment costs (number of nodes and communication lines).

Key words: parametrically defined regular graph, average distance, parametric description, circulant network, optimal graph, large language model.

References

Bibliographic reference: Monakhov O. G., Monakhova E. A. Construction and Optimization of Topological Structures of Regular Networks with Parametric Description //"Problems of informatics",  2026, №3, pp.60-74.  DOI: 10.24412/2073-0667-2026-3-60-74.

V. A. Perepelkin^1,3 , I. I. Titov2 , N. L. Podkolodny^1,2 , N. E. Matus^1,3 , E. I. Shevelev³ , N. A. Kolchanov² , M. A. Marchenko^1,3

1 Institute of Computational Mathematics and Mathematical Geophysics SB RAS, 630090, Novosibirsk, Russia

2 Institute of Cytology and Genetics SB RAS, 630090, Novosibirsk, Russia

3 Novosibirsk State University, 630090, Novosibirsk, Russia

ACTIVE KNOWLEDGE BASE CONSTRUCTION FOR AUTOMATIC ANALYSIS OF SECONDARY RNA STRUCTURE

The paper addresses the problem of automating the solution of bioinformatics problems by using active knowledge bases, which provide a partial formal description of a problem domain together with a library of existing software modules. Modern bioinformatics deals with rapidly growing volumes of genetic data and a large number of independently developed analysis tools, so manual search and integration of these tools becomes time-consuming and error-prone. Unlike widely used workflow systems that require users to manually compose computational pipelines, the proposed approach constructs a solution program automatically from a high-level specification of the task. The active knowledge base contains a computational model represented as a bipartite directed graph. The vertices of this graph are variables and operations. Variables describe quantities of the problem domain, and operations describe possible ways of computing some variables from other variables. Each operation is linked to a software module that implements it. These modules are annotated with functional properties, such as input and output data types, and with non-functional properties, such as execution time, memory usage, accuracy, and suitability for particular data characteristics. When a user specifies input and output variables, the system searches the computational model for a subgraph that connects the given inputs to the required outputs. This subgraph defines a problem solution algorithm. The system then selects appropriate modules for each operation, orders their execution, and automatically runs independent operations in parallel. If several candidate solutions exist, the system is able to choose the one that best satisfies user-defined non-functional criteria.

This approach is applied to the problem of RNA secondary structure analysis. In this problem, the input is a set of RNA sequences, and the outputs are the predicted secondary structures. The secondary structure of an RNA molecule is defined as a set of complementary base pairs that satisfies two constraints: each position can participate in at most one pair, and pseudoknots are forbidden. The problem is formulated as a discrete optimization problem. For a single sequence, the goal is to find the structure with minimum free energy. For a set of aligned sequences, the goal is to predict a common secondary structure from the alignment. The active knowledge base includes variables for primary RNA sequences, aligned sequences, secondary structures, and a common secondary structure.

It includes operations for predicting a secondary structure from a single sequence, for performing multiple sequence alignment, and for predicting a common secondary structure from aligned sequences. The computational model contains two branches. One branch applies the single-sequence prediction operation independently to each input sequence. The other branch first performs multiple alignment and then predicts a common structure. A more detailed model also includes splitting and merging operations, which allow the system to process independent subsets of sequences in parallel. Therefore, the system can automatically construct not only a sequential solution but also a parallel solution adjusted to available computational resources.

A prototype of the active knowledge base was developed using existing modules from the ViennaRNA package. The prototype was implemented on the Aka subsystem of the HPC Community Cloud platform. Testing was performed on 58 nucleotide sequences of evolutionarily conserved domains 5 and 6 of group II intron from organelle genomes of Solanoideae. The automatically constructed pipeline followed the branch that performs multiple sequence alignment and then predicts the common secondary structure. The analysis showed that domain 5 is shorter and more conserved than domain 6, while the fraction of nucleotides involved in complementary interactions is similar in both domains. The predicted secondary structures were consistent with accepted structural models of these domains. This confirms the correctness of the automatically constructed solution.

The results show that the active knowledge approach hides low-level programming and manual module integration from the user. A specialist only needs to formulate a high-level task, while the system automatically finds, combines, and executes suitable software modules. The same active knowledge base can be reused for different tasks within the domain. In the future, the approach can be extended by including more modules and by integrating with existing bioinformatics platforms. This would allow such platforms to store not only tools and technical descriptions, but also expert knowledge about module properties and usage practices, thereby improving the reliability and adequacy of automatically constructed solutions.

Key words: bioinformatics, automatic program construction, secondary RNA structure analysis, active knowledge concept.

This work was carried out under state contract with ICMMG SB RAS FWNM-2025-0005 and supported by the budget project No. FWNR-2026-0023 of the Federal Research Center ICG SB RAS.

References

Bibliographic reference: Perepelkin V. A., Titov I. I., Podkolodny N. L., Matus N. E., Shevelev E. I., Kolchanov N. A., Marchenko M. A. Active Knowledge Base Construction for Automatic Analysis of Secondary RNA Structure//"Problems of informatics",  2026, №3, pp..75-94.  DOI: 10.24412/2073-0667-2026-3-75-94.