Wind provides clean, renewable energy. The core concept is simple: wind turbines spin blades to generate power. However, today's systems are anything but simple. Modern wind turbines have blades that sweep a 120 meter circle, cost more than 1 million dollars and generate multiple megawatts of power. Each turbine may include up to 1,000 sensors and actuators – integrating strain gages, bearing monitors and power conditioning technology. The turbine can control blade speed and power generation by altering the blade pitch and power extraction. Controlling the turbine is a sophisticated job requiring many cooperating processors closing high-speed loops and implementing intelligent monitoring and optimization algorithms. But the real challenge is integrating these turbines so that they work together. A wind farm may include hundreds of turbines. They are often installed in difficult-to-access locations at sea. The farm must implement a fundamentally and truly distributed control system. Like all power systems, the goal of the farm is to match generation to load. A farm with hundreds of turbines must optimize that load by balancing the loading and generation across a wide geography. Wind, of course, is dynamic. Almost every picture of a wind farm shows a calm sea and a setting sun. But things get challenging when a storm goes through the wind farm. In a storm, the control system must decide how to take energy out of gusts to generate constant power. It must intelligently balance load across many turbines. And a critical consideration is the loading and potential damage to a half-billion-dollar installed asset. This is no environment for a slow or undependable control system. Reliability and performance are crucial.

General Atomics had to meet strict requirements for data bandwidth and availability. They could have designed and implemented a custom solution for data integration on their own, but that would have significantly delayed the project. Plus the lifecycle cost of developing and maintaining a custom software solution for a specific project can be enormous.

In research programs, the compute intensity of any specific module cannot be determined ahead of time and is subject to continuous change; the ability of one compute engine to manage its load cannot be determined until runtime. The infrastructure needs to provide clear feedback when communication deadlines are not being met byconnected modules. This requires the ability to rapidly and easily reallocate modules to networked nodes. In other words, modules should be able to be reorganized around the distributed architecture without the time and effort of reconfiguring the underlying network integration. This way, researchers can stay focused on application level issues rather than dealing with system architecture problems created by changes in systems integration.