How Rakuten fuels speed, performance, and scale with Aerospike

Rakuten is a global leader in internet services that empower individuals, communities, businesses, and society. One of their major revenue streams comes from advertising to shoppers. Through their display advertising platform, Rakuten provides different kinds of non-targeted and targeted ads to support e-commerce, enabling merchants to leverage their media ecosystem to promote the right products to the right audience.

The necessity of a high-performance database to handle massive volume and load

In Rakuten’s advertising platform, users enter the internet shopping mall, Rakuten Ichiba, through a front-end application or mobile app. A tracking database collects information about the user from cookies or a mobile phone ID, which forwards this information to a recommendation engine and user mapping engine. These engines send user profile data and user-specific recommendations to a targeting database, which stores the information and forwards it to an ad delivery engine to be served up when a user visits a website.

To be effective, the targeting database must be able to store enormous amounts of recommendation and user mapping data while scaling to meet ever-increasing traffic demands and providing round-the-clock availability. In addition, whenever Rakuten serves ads to users from the internet, demand can grow dramatically. That’s because targeting a massive number of visitors to a given website means that computation by each machine learning model can double or triple.

Prior to selecting Aerospike, Rakuten’s databases were not fully optimized for their targeting ad platform. The Rakuten team knew they needed a data store to handle the speed and volume required to support their targeted ads service.

Specifically, they needed a NoSQL database that met several critical requirements:

• Large data store

  The database would need to handle massive amounts of recommendation and user mapping data for millions of individual user records, as well as consider traffic fluctuations depending on time of day and month, specific event, and more.

• Low latency

  The database architecture would have to provide low latency for Rakuten’s extremely high-volume data environment. It could not be a memory-first, or “cache,” solution that relied on RAM to store data because the company’s ad data is so voluminous.

• Unlimited scalability

  Rakuten’s continually growing volume of data and user traffic requires a database with unlimited scalability. The team would also need the ability to easily scale the database by adding nodes (such as servers).

• High availability

  Ad services must be up and running 24/7 because any downtime means lost revenue. The database would need to be highly available, allowing the Rakuten team to maintain and upgrade it without stopping the service or using many resources.

• Multi-model

  Rakuten’s machine learning teams are always introducing new models. In light of this, the database would have to support multiple data models to ingest the different kinds of data sources required to achieve the desired output based on specific training models.

Multi-model database supports innovation across the enterprise

By using Aerospike as their targeting database to house user profiles for ad targeting, Rakuten now has a clearer view of their source data, such as browsing, purchase, demographics, user mapping, and certificate. This data can be shared throughout the company, including with developers, to do things like create new products and develop new algorithms.

In addition, the Aerospike database enables Rakuten to optimize advertising by using their own recommendation data to create and train machine learning models and make predictions about what users will likely react to based on their buying behavior. They can also use it to evaluate whether they display ads to the right users and to monitor for performance issues.

Watch our video to learn more about Rakuten’s advertising platform and how they used Aerospike to help boost revenue.

• Seamless scaling with no performance loss

  Aerospike scaled linearly to 2TB while maintaining peak performance

• Improved performance with lower costs and higher revenue

  Aerospike boosted performance by reducing TCO, increasing revenue, and streamlining infrastructure

• Automated issue resolution safeguards server clusters

  Automating server issue resolution protects clusters and optimizes resource usage

• Enhanced machine learning supports better ad targeting

  Multi-model support enables advanced machine learning and more effective ad targeting