Nippon Telegraph and Telephone : Demonstration of the practicability of a secure computation environment utilizing low latency communication of the IOWN All Photonics Network Realize the integration of secure computation servers at multiple locations

News Highlights:

1. ●In a secure computation system using secret sharing, the common practice was to place all servers on a single data center to minimize the impact of a large volume of telecommunications
2. ●By utilizing data centers interconnected by APN with low latency, we confirmed that data analysis by secure computation is possible in a practical time even in a computing environment distributed in multiple data centers
3. ●From the viewpoint of disaster countermeasures, energy optimization, and security, it is expected to provide safe and secure data utilization by multiple organizations (e.g., the construction of high-precision AI models by utilizing data centers distributed in multiple locations)

Tokyo - June 12, 2024 - NTT Corporation (Headquarters: Chiyoda Ward, Tokyo; Representative Member of the Board and President: Akira Shimada; hereinafter "NTT") has demonstrated that data analysis by secure computation can be achieved in a practical amount of time even in a computing environment distributed in multiple data centers by utilizing high capacity and low latency communication - IOWN All Photonics Network (APN¹). This achievement is expected to create a platform that enables secure data utilization by connecting a secure computation server located at distant data centers with IOWN APN.

1. Background

NTT has been conducting research on a wide range of social values, information utilization, cyber security, privacy, ethics, laws, systems, etc., with the aim of transforming and developing advanced social systems and human societies through information and communications technology (ICT). To appropriately manage data and promote its utilization, we have been working for many years on the development of secure computation technology that uses secret sharing, and our proposed computation method has been adopted as the international standard by the International Organization for Standardization (ISO). ^2,3
In addition, with the increasing demand for computing functions such as AI and IoT, from the viewpoint of disaster countermeasures, energy optimization, and security, in addition to a single giant data center, a combination of data centers distributed in multiple locations is expected to be utilized. In a conventional secure computation system using secret sharing, while high-speed analysis is possible, all servers are generally placed in a single data center to minimize the impact of a large volume of communications. In the newly proposed method, to utilize secure computation by data centers distributed in multiple locations, we interconnected data centers using APN - a high-capacity and low-latency communication network - and accumulated techniques and use-case studies.

Figure 1 Secure Computation System

2. Details of Initiatives

To realize secure data utilization across a wider area and among multiple organizations, we have built a secure computation system with distributed servers for distributed IOWN APN data centers built by NTT WEST. To evaluate the impact of the network during computation processing, we measured and compared the computation performance of the IOWN APN connection configuration, the configuration simulating a general network connection, and the configuration in a single data center.

From the result, we found a significant difference in the learning process (GBDT⁴ and FFNN⁵) of the AI model using 100,000 dummy data sets as learning targets. For FFNN, we confirmed that IOWN APN completed learning in approx. 1/7of the time⁶ compared to general network connection. We also compared the processing performance of a single data center with data centers distributed in multiple locations and confirmed that the processing time⁶ was approx. 3/2 times⁶ which is sufficient for practical use.
This result shows that it is possible to construct a secure computation system by connecting data centers at a fixed distance, which was limited to the same single data center due to communication speed and delay, and it is expected to be applied to a secure data utilization system between different data centers and enterprises.

Figure 2 Outline of Experiment Configuration

3. Outlook

In this demonstration experiment, we proved that a single secure computation system across distributed data centers can be realized by using IOWN APN. Going forward, we will continue to develop technologies through joint experiments to verify use cases for value creation by applying IOWN technology, such as "safe and secure data utilization across multiple businesses" and "utilization of computing resources by integrated regional data centers," and to solve operational issues for the implementation.

¹IOWN APN (All Photonics Network)
IOWN consists of three main components: the All-Photonics Network (APN,) which uses optical processing for not only networks but also device terminal processing; Digital Twin Computing, which enables advanced real-time interaction between things and people in cyberspace; and the Cognitive Foundation, which efficiently deploys various ICT resources including these.
By introducing new optical technologies into the network, terminals, and chips, APN achieves ultra-low power consumption and ultra-high-speed processing, which were previously difficult to achieve. By allocating wavelengths for each function on a single optical fiber, it is possible to provide multiple functions that support social infrastructure, such as information communication functions such as the Internet and sensing functions, without interfering with each other.
https://www.rd.ntt/e/iown/

²Contribution to the first ISO for Secure Computation Technology for data utilization and privacy protection
- Leading the ISO standardization through many years of R&D knowledge - (September 2023)
https://group.ntt/en/newsrelease/2023/09/15/230915a.html

³NTT's Secure Computing Technology adopted as ISO International Standard
- Secure computing technology to speed up data utilization and protect privacy - (March 2024)
https://group.ntt/en/newsrelease/2024/03/21/240321b.html

⁴GBDT (Gradient Boosting Decision Tree):
A machine learning model that combines multiple decision trees, and a data analysis method using tree structures (tree diagrams,) to make more advanced predictions.

⁵FFNN (Feedforward Neural Network):
A type of neural network (NN) that combines neurons into a layered network. The final result is derived by propagating values through three types of layers: an input layer, a hidden layer, and an output layer.

⁶FFNN model learning time using 100,000 dummy data sets:
APN connection (approx. 22 minutes), general network simulation (approx.157 minutes), single data center (approx. 15 minutes).

About NTT

NTT contributes to a sustainable society through the power of innovation. We are a leading global technology company providing services to consumers and businesses as a mobile operator, infrastructure, networks, applications, and consulting provider. Our offerings include digital business consulting, managed application services, workplace and cloud solutions, data center and edge computing, all supported by our deep global industry expertise. We are over $97B in revenue and 330,000 employees, with $3.6B in annual R&D investments. Our operations span across 80+ countries and regions, allowing us to serve clients in over 190 of them. We serve over 75% of Fortune Global 100 companies, thousands of other enterprise and government clients and millions of consumers.