JD.com's Archimedes: A Comprehensive Data Center Operating System

JD.com's Archimedes is a comprehensive data center operating system that provides one-click scaling, intelligent scheduling, mixed deployment, and multi-site active-active capabilities. It can manage applications of any scale, from a few nodes to tens of thousands of nodes, and provide elastic expansion for the entire cluster to easily cope with the explosive growth of business.

Archimedes uses machine learning technology to intelligently schedule applications, combining application resource requests and historical resource usage to dynamically deploy and schedule applications. It can also mix deploy different types of big data applications, such as Hadoop, Spark, Kafka, Hbase, Storm, and Flink, in the same container cloud cluster to improve resource utilization.

Archimedes adopts a wide-area distributed architecture, connecting multiple data centers in different locations into a whole, and realizing transparent data synchronization and fault switching through built-in middleware container cloud services. It supports multi-data center application deployment and scheduling, and realizes closed-loop deployment of core applications, effectively improving system high availability and solving data center-level scalability issues.

During the 2017 Double 11 shopping festival, Archimedes had already taken over application resource scheduling, scheduling millions of container instances daily and providing up to 30 million core-hours of computing resources for offline computing, with a service fulfillment rate of 98.3%. It has significantly reduced IT costs for JD.com and will continue to open up and empower more customers in the era of boundaryless retail.