How Fujitsu implemented a global data mesh architecture and democratized data

This is a guest post co-authored with Kanehito Miyake, Engineer at Fujitsu Japan. 

Fujitsu Limited was established in Japan in 1935. Currently, we have approximately 120,000 employees worldwide (as of March 2023), including group companies. We develop business in various regions around the world, starting with Japan, and provide digital services globally. To provide a variety of products, services, and solutions that are better suited to customers and society in each region, we have built business processes and systems that are optimized for each region and its market.

However, in recent years, the IT market environment has changed drastically, and it has become difficult for the entire group to respond flexibly to the individual market situation. Moreover, we are challenged not only to revisit individual products, services, and solutions, but also to reinvent entire business processes and operations.

To transform Fujitsu from an IT company to a digital transformation (DX) company, and to become a world-leading DX partner, Fujitsu has declared a shift to data-driven management. We built the OneFujitsu program, which standardizes business projects and systems throughout the company, including the domestic and overseas group companies, and tackles the major transformation of the entire company under the program.

To achieve data-driven management, we built OneData, a data utilization platform used in the four global AWS Regions, which started operation in April 2022. As of November 2023, more than 200 projects and 37,000 users were onboarded. The platform consists of approximately 370 dashboards, 360 tables registered in the data catalog, and 40 linked systems. The data size stored in Amazon Simple Storage Service (Amazon S3) exceeds 100 TB, including data processed for use in each project.

In this post, we introduce our OneData initiative. We explain how Fujitsu worked to solve the aforementioned issues and introduce an overview of the OneData design concept and its implementation. We hope this post will provide some guidance for architects and engineers.

Challenges

Like many other companies struggling with data utilization, Fujitsu faced some challenges, which we discuss in this section.

Siloed data

In Fujitsu’s long history, we restructured organizations by merging affiliated companies into Fujitsu. Although organizational integration has progressed, there are still many systems and mechanisms customized for individual context. There are also many systems and mechanisms overlapping across different organizations. For this reason, it takes a lot of time and effort to discover, search, and integrate data when analyzing the entire company using a common standard. This situation makes it difficult for management to grasp business trends and make decisions in a timely manner.

Under these circumstances, the OneFujitsu program is designed have one system per one business globally. Core systems such as ERP and CRM are being integrated and unified in order to not have silos. It will make it easier for users to utilize data across different organizations for specific business areas.

However, to spread a culture of data-driven decision-making not only in management but also in every organization, it is necessary to have a mechanism that enables users to easily discover various types of data in organizations, and then analyze the data quickly and flexibly when needed.

Excel-based data utilization

Microsoft Excel is available on almost everyone’s PC in the company, and it helps lower the hurdles when starting to utilize data. However, Excel is mainly designed for spreadsheets; it’s not designed for large-scale data analytics and automation. Excel files tend to contain a mixture of data and procedures (functions, macros), and many users casually copy files for one-time use cases. It introduces complexity to keep both data and procedures up to date. Furthermore, it tends to require domain-specific knowledge to manage the Excel files for individual context.

For those reasons, it was extremely difficult for Fujitsu to manage and utilize data at scale with Excel.

Solution overview

OneData defines three personas:

• Publisher – This role includes the organizational and management team of systems that serve as data sources. Responsibilities include:
  • Load raw data from the data source system at the appropriate frequency.
  • Provide and keep up to date with technical metadata for loaded data.
  • Perform the cleansing process and format conversion of raw data as needed.
  • Grant access permissions to data based on the requests from data users.
• Consumer – Consumers are organizations and projects that use the data. Responsibilities include:
  • Look for the data to be used from the technical data catalog and request access to the data.
  • Handle the process and conversion of data into a format suitable for their own use (such as fact-dimension) with granted referencing permissions.
  • Configure business intelligence (BI) dashboards to provide data-driven insights to end-users targeted by the consumer’s project.
  • Use the latest data published by the publisher to update data as needed.
  • Promote and expand the use of databases.
• Foundation – This role encompasses the data steward and governance team. Responsibilities include:
  • Provide a preprocessed, generic dataset of data commonly used by many consumers.
  • Manage and guide metrics for the quality of data published by each publisher.

Each role has sub-roles. For example, the consumer role has the following sub-roles with different responsibilities:

• Data engineer – Create data process for analysis
• Dashboard developer – Create a BI dashboard
• Dashboard viewer – Monitor the BI dashboard

The following diagram describes how OneData platform works with those roles.

Let’s look at the key components of this architecture in more detail.

Publisher and consumer

In the OneData platform, the publisher is per each data source system, and the consumer is defined per each data utilization project. OneData provides an AWS account for each.

This enables the publisher to cleanse data and the consumer to process and analyze data at scale. In addition, by properly separating data and processing, it becomes effortless for the teams and organizations to share, manage, and inherit processes that were traditionally confined to individual PCs.

Foundation

When the teams don’t have a robust enough skillset, it can require more time to model and process data, and cause longer latency and lower data quality. It can also contribute to lower utilization by end-users. To address this, the foundation role provides an already processed dataset as a generic data model for data commonly use cases used by many consumers. This enables high-quality data available to each consumer. Here, the foundation role takes the lead in compiling the knowledge of domain experts and making data suitable for analysis. It is also an effective approach that eliminates duplicates for consumers. In addition, the foundation role monitors the state of the metadata, data quality indicators, data permissions, information classification labels, and so on. It is crucial in data governance and data management.

BI and visualization

Individual consumers have a dedicated space in a BI tool. In the past, if users wanted to go beyond simple data visualization using Excel, they had to build and maintain their own BI tools, which caused silos. By unifying these BI tools, OneData lowers the difficulty for consumers to use BI tools, and centralizes operation and maintenance, achieving optimization on a company-wide scale.

Additionally, to keep portability between BI tools, OneData recommends users transform data within the consumer AWS account instead of transforming data in the BI tool. With this approach, BI tool loads data from AWS Glue Data Catalog tables through an Amazon Athena JDBC/ODBC driver without any further transformations.

Deployment and operational excellence

To provide OneData as a common service for Fujitsu and group companies around the world, Regional OneData has been deployed in several locations. Regional OneData represents a unit of system configurations, and is designed to provide lower network latency for platform users, and be optimized for local languages, working hours for system operations and support, and region-specific legal restrictions, such as data residency and personal information protection.

The Regional Operations Unit (ROU), a virtual organization that brings together members from each region, is responsible for operating regional OneData in each of these regions. OneData HQ is responsible for supervising these ROUs, as well as planning and managing the entire OneData.

In addition, we have a specially positioned OneData called Global OneData, where global data utilization spans each region. Only the properly cleansed and sanitized data is transferred between each Regional OneData and Global OneData.

Systems such as ERP and CRM are accumulating data as a publisher for Global OneData, and the dashboards for executives in various regions to monitor business conditions with global metrics are also acting as a consumer for Global OneData.

Technical concepts

In this section, we discuss some of the technical concepts of the solution.

Large scale multi-account

We have adopted a multi-account strategy to provide AWS accounts for each project. Many publishers and consumers are already onboarded into OneData, and the number is expected to increase in the future. With this strategy, future usage expansion at scale can be achieved without affecting the users.

Also, this strategy allowed us to have clear boundaries in security, costs, and service quotas for each AWS service.

All the AWS accounts are deployed and managed through AWS Organizations and AWS Control Tower.

Serverless

Although we provide independent AWS accounts for each publisher and consumer, both operational costs and resource costs would be enormous if we accommodated individual user requests, such as, “I want a virtual machine or RDBMS to run specific tools for data processing.” To avoid such continuous operational and resource costs, we have adopted AWS serverless services for all the computing resources necessary for our activities as a publisher and consumer.

We use AWS Glue to preprocess, cleanse, and enrich data. Optionally, AWS Lambda or Amazon Elastic Container Service (Amazon ECS) with AWS Fargate can also be used based on preferences. We allow users to set up AWS Step Functions for orchestration and Amazon CloudWatch for monitoring. In addition, we provide Amazon Aurora Serverless PostgreSQL as standard for consumers, to meet their needs for data processing with extract, load, and transform (ELT) jobs. With this approach, only the consumer who requires those services will incur charges based on usage. We are able to take advantage of lower operational and resource costs thanks to the unique benefit of serverless (or more accurately, pay-as-you-go) services.

AWS provides many serverless services, and OneData has integrated them to provide scalability that allows active users to quickly provide the required capability as needed, while minimizing the cost for non-frequent users.

Data ownership and access control

In OneData, we have adopted a data mesh architecture where each publisher maintains ownership of data in a distributed and decentralized manner. When the consumer discovers the data they want to use, they request access from the publisher. The publisher accepts the request and grants permissions only when the request meets their own criteria. With the AWS Glue Data Catalog and AWS Lake Formation, there is no need to update S3 bucket policies or AWS Identity and Access Management (IAM) policies every time we allow access for individual data on an S3 data lake, and we can effortlessly grant the necessary permissions for the databases, tables, columns, and rows when needed.

Conclusion

Since the launch of OneData in April 2022, we have been persistently carrying out educational activities to expand the number of users and introducing success stories on our portal site. As a result, we have been promoting change management within the company and are actively utilizing data in each department. Regional OneData is being rolled out gradually, and we plan to further expand the scale of use in the future.

With its global expansion, the development of basic functions as a data utilization platform will reach a milestone. As we move forward, it will be important to make sure that OneData platform is used effectively throughout Fujitsu, while incorporating new technologies related to data analysis as appropriate. For example, we are preparing to provide more advanced machine learning functions using Amazon SageMaker Studio with OneData users and investigating the applicability of AWS Glue Data Quality to reduce the manual quality monitoring efforts. Furthermore, we are currently in the process of implementing Amazon DataZone through various initiatives and efforts, such as verifying its functionality and examining how it can operate while bridging the gap between OneData’s existing processes and to the ideal process we are aiming for ideals.

We have had the opportunity to discuss data utilization with various partners and customers and although individual challenges may differ in size and its context, the issues that we are currently trying to solve with OneData are common to many of them.

This post describes only a small portion of how Fujitsu tackled challenges using the AWS Cloud, but we hope the post will give you some inspiration to solve your own challenges.

About the Author

Junpei Ozono