MMC MSU: Network Powered by Computing (NPC)

Main article: Virtualization. Classification and applications

2025: Network Powered by Computing Presentation

Representatives of the NOSH "Brain, Cognitive Systems, Artificial Intelligence" proposed a new generation computing infrastructure architecture based on the use of machine learning methods and multi-agent systems. This approach allows you to manage resource allocation across scalable network environments for high performance, resiliency, and security. Lomonosov Moscow State University - Faculty of Computational Mathematics and Cybernetics (VMK MSU) announced this on June 4, 2025.

The rapid growth of a variety of applications, the volume of data to be processed, has significantly exacerbated the need for computing. Today's applications are becoming increasingly demanding on computing resources, data transfer rates, and scalability. While traditional data centers (data centers) are capable of providing computing services only on their local infrastructure, new calls require distributed solutions with global connectivity and high adaptability. In other words, the on-demand calculation service is required. No matter where and which computers will process the data, it is important to quickly get the result.

A striking example of this is the problem of creating a computing infrastructure for training large models in the field of artificial intelligence. LLM models have already reached such a size that they do not fit into one. DPC They need a network of data centers. Another example is the problem of controlling the distribution of computing power. states As of June 2025, computing power is a kind of analogue of power power. All economically developed countries have automated power distribution management systems. The problem of creating and managing a single network of computing power is one of the key challenges for. RUSSIAN FEDERATION

In response to this challenge, scientists at VMK Moscow State University have proposed a computing infrastructure architecture called Network Powered by Computing (NPC), which provides flexibility, scalability and efficiency in managing computing resources in wide area networks for on-demand computing. The architecture is based on the use of machine learning methods and multi-agent systems, which allow you to dynamically control the distribution of tasks between computers depending on the load and requirements of applications.

NPC is a fully virtualized and software-managed infrastructure that integrates various computing installations and high-speed data networks. An important feature is that such an infrastructure is not limited to one data center or computing complex. It combines heterogeneous computing resources into a single ecosystem, giving users the ability to get computing power on demand, anywhere, anytime.

The main properties of NPC include:

• Global connectivity and resource availability, allowing users to leverage computing power without being tied to specific data centers.
• Deterministic quality of data services, providing predictable latency and stable performance.
• Virtualization, scalability, and serverlessness to dynamically reconfigure resources based on workload.
• Availability, reliability, and resiliency through resource duplication and intelligent data flow management.
• Efficient resource allocation using machine learning techniques that take into account current load and consumption forecasts.
• Security through tiered data protection and network connectivity.

One of the key features of NPC is the use of machine learning techniques to predict load and optimal resource allocation. For example, machine learning allows you to predict the execution time of applications on different computers, optimally distribute data streams in the network and select the best data transmission channels.

Multi-agent systems are used to control the distribution of tasks between computers. In these systems, agents communicate with each other by exchanging information about the load and state of resources. This allows you to dynamically adapt the architecture to changes in application requirements and network infrastructure.

NPC opens up opportunities for the development of cloud computing, high-performance computing (HPC) installations, telecommunications and Internet services. This is especially true for applications that require high performance, such as artificial intelligence, machine learning, big data analysis, and augmented reality systems.

Virtualization and software resource management enable you to leverage computing power more efficiently, reducing hardware and maintenance costs. In addition, the architecture allows you to integrate computing resources from various geographical locations, creating global distributed systems.

Our architecture of computing infrastructure - NPC allows you to more efficiently manage computing resources in scalable network environments, providing the quality of computing service required by the user. The use of machine learning methods and multi-agent systems provides dynamic task distribution and adaptation to changing application requirements. This opens up new prospects for the creation of new generation computing infrastructures, - said Ruslan Smelyansky, head of the Department of Computational Mathematics and Cybernetics, VMK Moscow State University.

Future research will focus on integrating NPC with artificial intelligence systems and creating hybrid architectures that combine cloud and peripheral computing. It is also planned to study the issues of security and data protection in distributed computing environments.