Real-time control of Extremely Large Telescope mirror systems using on-line high performance computing

Many Big Physics applications require an enormous amount of computational power to solve large problems with demanding real-time constraints. Typically, a large number of data channels acquired in real-time feed mathematical routines that generate outputs that act on real-world processes. The number of sensor and actuator channels can be in the 100-100,000 range, and the mathematical routines can be as sophisticated as real-time solving of non-linear partial differential equations. The time constraints are often in the 1 ms range, or faster, per cycle. The most obvious solution to such a catalog of requirements is programming highly specialized hardware to meet the goals. The drawback with this approach is that, by the time the final system is deployed, one loses almost all connections to the original design and mathematical context. Even slight changes regarding those components translate into weeks of reprogramming. We report on an alternative approach based on the National Instruments Lab VIEW graphical programming environment. The test cases are the mirror control systems of the European Southern Observatory (ESO) European Extremely Large Telescope (E-ELT) currently in the design and proof of-concept phase. We explain how the entire control system development cycle consisting of design, simulation, and real-time deployment can be completed without sacrificing the high-level description of the system. This approach allows fundamental changes in design and mathematical implementation without costly reprogramming and redeployment efforts. The required computational power is provided by multi-core implementations on standard commercial off-the-shelf (COTS) hardware. Several benchmarks illustrate what can be achieved with such an approach.