Advanced technologies, powerful alliances and open collaboration — this is the YADRO recipe for the rapid progress, proven by practice. Enjoy the taste of this mix, just like we do.
Why we can
Great things are created by great people. It takes a special company to accumulate and nurture talents. Become familiar with YADRO culture and become a fan or a valued team member.
Who we are
Мы открываем, конструируем, разрабатываем и создаем. Совершаем прорывы в технологиях корпоративного уровня. Становимся источником инноваций. Создаем выдающиеся продукты для современных технологических задач.
Наше призвание
TATLIN.UNIFIED
Передовые технологии, мощные стратегические партнерства и открытая коллаборация являются рецептом YADRO для достижения прорывов в сфере высоких технологий, доказанных на деле.
Что дает нам импульс к прогрессу
Все прекрасное в мире создается талантливыми людьми. Для сплочения и взращивания талантов требуется особая команда. Познакомьтесь с культурой YADRO и станьте ценителем или неотъемлемой частью нашей команды.
О нашей комнанде

Leveraging AI power to predict storage system failures, performance degradation, data unavailability and loss

Superior Storage Reliability Driven by Machine Learning

YADRO joint research effort together with
the National Research University Higher School of Economics (HSE) and Peter the Great St. Petersburg Polytechnic University
Today, in the era of big data, IoT and real-time analytics, enterprises need to permanently store an ever-growing volume of structured and unstructured data. According to IDC forecast, amounts of data stored worldwide will double in the next few years. At the same time, data availability and reliability requirements are constantly increasing. It is critical to minimize the risk of a sudden data damage, loss or breach. Data access should be easy, lightning-fast, continuous, and predictable.

While storage infrastructure requirements are constantly evolving, failures can become quite costly for businesses. Accurate and timely prediction of failures in a storage system enables enterprises to become proactive in preventing data unavailability or loss.
Our innovative approach is based on a combination of several deep neural networks and designed to predict abnormal behavior of TATLIN platform. The idea behind this approach is to focus AI on identifying patterns that are far ahead of our engineering thinking, as well as scenarios that are far beyond implemented platform reliability and redundancy requirements. Powered by machine learning to process historical health data and enable monitoring in real time, TATLIN platform became capable of predicting various types of non-obvious storage service failures that might impact performance and data availability, and result in data loss.

To achieve a higher level of accuracy when making such predictions, TATLIN platform utilizes hybrid interpretation of real and simulated storage health data, which is one of the key innovative results of this research project.
Today, in the era of big data, IoT and real-time analytics, enterprises need to permanently store an ever-growing volume of structured and unstructured data. According to IDC forecast, amounts of data stored worldwide will double in the next few years. At the same time, data availability and reliability requirements are constantly increasing. It is critical to minimize the risk of a sudden data damage, loss or breach. Data access should be easy, lightning-fast, continuous, and predictable.

While storage infrastructure requirements are constantly evolving, failures can become quite costly for businesses. Accurate and timely prediction of failures in a storage system enables enterprises to become proactive in preventing data unavailability or loss.

We believe that it is extremely important for TATLIN platform to provide reliable, redundant and highly available storage services. To enable this goal, one of our joint research initiatives focused on finding an effective solution to mitigate TATLIN platform failure and data loss risks by leveraging AI-powered algorithms.

We've partnered with National Research University Higher School of Economics (HSE) and Peter the Great St. Petersburg Polytechnic University (SPbPU) to develop AI-powered algorithms in order to enable predictive storage platform maintenance and health monitoring. As a result of our joint research effort, we have developed an enterprise-ready built-in risk mitigation functionality for TATLIN storage platform that leverages homegrown AI algorithms.
We believe that it is extremely important for TATLIN platform to provide reliable, redundant and highly available storage services. To enable this goal, one of our joint research initiatives focused on finding an effective solution to mitigate TATLIN platform failure and data loss risks by leveraging AI-powered algorithms.

We've partnered with National Research University Higher School of Economics (HSE) and Peter the Great St. Petersburg Polytechnic University (SPbPU) to develop AI-powered algorithms in order to enable predictive storage platform maintenance and health monitoring. As a result of our joint research effort, we have developed an enterprise-ready built-in risk mitigation functionality for TATLIN storage platform that leverages homegrown AI algorithms.
Offering built-in self-diagnostic capabilities powered by machine learning, TATLIN platform is able to automatically predict critical situations such as performance degradation, data loss, data path failure, and minimize other risks.
As part of the effort, our joint research team developed a digital storage twin of YADRO TATLIN platform with reinforced learning capabilities which produced diverse failure scenarios much faster than the real-life storage platform. This approach helped to educate AI-powered algorithms with better-quality input data. In addition, our team used a large fabric of ML algorithms, both supervised and unsupervised, to make more accurate and robust predictions.

Today, TATLIN platform includes an early version of predictive maintenance features that are going to gain more analytical power as runtime continues to grow.

We believe in the importance of innovation in predictive maintenance and plan to continue our joint research effort in this area. Later this year, we plan to publish scientific papers and patents related to this research.
Our innovative approach is based on a combination of several deep neural networks and designed to predict abnormal behavior of TATLIN platform. The idea behind this approach is to focus AI on identifying patterns that are far ahead of our engineering thinking, as well as scenarios that are far beyond implemented platform reliability and redundancy requirements. Powered by machine learning to process historical health data and enable monitoring in real time, TATLIN platform became capable of predicting various types of non-obvious storage service failures that might impact performance and data availability, and result in data loss.

To achieve a higher level of accuracy when making such predictions, TATLIN platform utilizes hybrid interpretation of real and simulated storage health data, which is one of the key innovative results of this research project. As part of the effort, our joint research team developed a digital storage twin of YADRO TATLIN platform with reinforced learning capabilities which produced diverse failure scenarios much faster than the real-life storage platform. This approach helped to educate AI-powered algorithms with better-quality input data. In addition, our team used a large fabric of ML algorithms, both supervised and unsupervised, to make more accurate and robust predictions.

Today, TATLIN platform includes an early version of predictive maintenance features that are going to gain more analytical power as runtime continues to grow.

We believe in the importance of innovation in predictive maintenance and plan to continue our joint research effort in this area. Later this year, we plan to publish scientific papers and patents related to this research.
United expertise
We amalgamated the deepest knowledge of AI, creative approach and technology proficiency. Academia and industry came together to devise a smart and efficient AI warden that is always working to make your TATLIN storage more reliable than you could hope for.
International Journal of Civil Engineering & Technology; 9, issue 11: 220-22, Article ID: IJCIET_09_11_022, publication Scopus indexed
A.V. Nevolin, YADRO, Moscow, Russia
M.E. Karpov, K. Arzymatov, V.S. Belavin, A.A. Sapronov and A.E. Ustyuzhanin, National Research University Higher School of Economics, Moscow, Russia
Published online at iaeme.com
Hybrid approach to design of storage attached network simulation system
Vladislav Belavin, Kenenbek Arzymatov, Maksim Karpov
National Research University Higher School of Economics, Moscow, Russia
Andrey Nevolin, YADRO, Moscow, Russia
Andrey Sapronov, Joint Institute for Nuclear Research, Dubna, Russia
Andrey Ustyuzhanin, Moscow Institute of Physics and Technology, Dolgoprudny, Russia
Published online at iaeme.com
Tuning hybrid distributed storage system digital twins by Reinforcement Learning
"Digital twins: artificial intelligence and classical modeling"
Vladislav Belavin, Kenenbek Arzymatov, Maksim Karpov, Andrey Sapronov, Andrey Ustyuzhanin
May 22, 2019
https://www.osp.ru/iz/ai2019/
AI Technologies Conference
"Digital twins reinforcement learning for preventive elimination of complex systems failures"
Vladislav Belavin, Kenenbek Arzymatov, Maksim Karpov, Andrey Sapronov, Andrey Ustyuzhanin
February 14, 2019
http://opentalks.ai/
OpenTalks.AI Conference