Linden Lab Case Study

Linden Lab is a San Francisco-based Internet company best known for its Second Life virtual world that provides a platform for users to generate and interact with 3D content. The Second Life virtual world can be accessed through Linden Lab’s client programs or via alternative third-party viewers. The company’s other offerings include Blocksworld, which lets users build and play with virtual 3D blocks, and “Project Sansar,” the code name for a new platform for virtual experiences that is scheduled for release in 2016.

• Company had created products that used immutable host images running on bare metal servers.
• The applications were monolithic and images took up to 15 minutes to deploy.
• For Project Sansar, the company chose to use Docker containers for increased speed and consistency between environments.
• Wanted to run Sansar at scale and needed efficient, fast cluster management and container orchestration.
• Initially evaluated a range of open-source tools, but encountered failures, immaturity, instability, and management difficulties.
  

• Started using Amazon EC2 Container Service (Amazon ECS), a scalable, high-performance container management service that provides cluster management and container orchestration. The company runs the containerized web applications and backend services of Project Sansar across clusters of Amazon EC2 instances.
• Built a custom scheduler for scheduling containers with specific PCI restrictions and CPU/memory requirements.
• Built a continuous integration and continuous delivery pipeline into Amazon ECS using Jenkins and Drone.io.
• Uses Elastic Load Balancing for automatic load balancing of traffic across containers and AWS CloudFormation to provision all Amazon ECS clusters and related resources.
  

• Amazon ECS eliminates the need to install software and operate servers; this reduces overall management needs for running containers at scale and helps speed time to market.
• Faster speed and efficiency in moving to containers, averaging one to two minutes for builds and one to three minutes for deployments—about 60 percent faster than previous build/deployment tools.
• Service is more stable and production ready.