Mobiuspace Delivers up to 40% Improved Price-Performance Using HAQM EMR on EKS

Founded in 2016, Mobiuspace is a global internet technology company that provides a diversified product portfolio for users to discover, explore, consume and create pan-entertainment content. This makes for a personalized experience anytime, anywhere.

With video streaming becoming the mainstay of mobile internet consumption, many users want to consume culturally-relevant content and find easier ways to access such information online. However, it was not easy, especially for users in Latin America and other emerging markets, to find localized and personalized content. Mobiuspace made it a priority to analyze and learn user behavior based on their media consumption, cultural, and national backgrounds to provide relevant video recommendations. This would lead to better localized and personalized video streaming services.

Its expanding services and customer base had also significantly driven up data operation costs. Its front-end server was processing as many as 100,000 QPS at peak hours and billions of users’ behavioral events. Mobiuspace wanted a cost-effective solution to address its massive data processing needs. It decided to improve the performance and efficiency of its big data operation by using AWS. This would help Mobiuspace keep pace with its rapid growth and boost business development through rapid cost reduction and continuous optimization.

Already running on HAQM EMR and HAQM Elastic Compute Cloud (HAQM EC2), Mobiuspace intended to better use these services to improve cluster resources utilization and gain more flexibility across AWS global infrastructure.

As the growing business placed increasing demands on its architecture, Mobiuspace underwent a data modernization effort and containerization transformation led by the big data team. Mobiuspace migrated its big data operation from HAQM EMR on EC2 to a fully-managed Kubernetes container platform—HAQM Elastic Kubernetes Service (HAQM EKS). With HAQM EMR on EKS, Mobiuspace integrated its big data and front-end applications to enable a microservice-based, containerized, and highly automated system and simpler operations and maintenance (O&M) management. In addition, HAQM EMR on EKS uses containers instead of virtual machines as the smallest resource unit to allow finer management and better utilization of resources.

For better virtual machine scheduling on HAQM EKS, Mobiuspace made full use of the AWS best practices: it runs Spot instances and HAQM EC2 instances powered by AWS Graviton processors to further reduce virtual machine costs of pod pools. HAQM EC2 Spot Instances allow users to tap into the unused EC2 capacity in the AWS Cloud. Available at up to a 90 percent discount compared to On-Demand prices, Spot instances are suitable for container and big data workloads. HAQM EMR or HAQM EKS also facilitate easy, seamless scheduling of and access to Spot resources. In 2020, HAQM EC2 instances powered by AWS Graviton processors were released. Mobiuspace’s testing on the containerized Java back-end services shows that HAQM EC2 M6g instances deliver 40 percent better price performance over M5 instances. “With HAQM EMR on EKS and the ARM-based AWS Graviton 2 instances, we improved the overall performance of our big data operations by 30 percent and reduced cost by 20 percent,” says Li Rui, vice president of technology at Mobiuspace.

Building on the modern data architecture of AWS, Mobiuspace uses HAQM SageMaker, a fully managed service that provides developers and data scientists with the ability to build, train, and deploy machine learning models quickly, to recommend video contents based on users’ interest. In addition, HAQM SageMaker is built with and optimizes commonly-used machine learning algorithms to save users from spending excessive time on algorithm selection and framework. Using HAQM SageMaker, Mobiuspace effectively shortened the cycles of continuous model iteration and updates to the optimized recommendation algorithm, improving user experience and customer satisfaction.