The STE||AR group will present two HPX demos at Supercomputing 2011 (LSU booth – 2839). These demos will highlight some of the capabilities any HPX application gains for free: active power management and automatic load balancing. If you are planning to visit Seattle for SC11 this year, please feel free to stop by. We will be happy to show you some of the possibilities of our parallel runtime system HPX.
Description of the two ParalleX demos for Supercomputing 2011:
- Adaptive Mesh Refinement and Active Power Management Demonstration
- Symmetric contact for impulsive loading computations
In this example, an entire adaptive mesh refinement simulation (AMR) evolving a hyperbolic system of equations using RK3 integration in time and second order spatial differencing is demonstrated.
Power usage is externally measured using the “Watt’s Up” power meter. Power usage is reported to the ipad interface:
The ipad application directly communicates with the HPX runtime system and can control the activity of the thread managers. A series of CPU kill-switches will toggle a particular thread manager associated with a specified core. Turning off a core will reduce the speed of the simulation but also reduce power consumption. The CPU clock speed can also be changed interactively through the ipad application.
A maximum power budget can be directly specified using the “Auto Power Management” tool in the upper right hand corner of the ipad application:
The HPX runtime system provides fine grained power management tools and performance counters that are independent of the application. This will be especially useful in situations where power budgets are constrained and vary in time.
In this example, various mesh objects consisting of different materials are interactively collided together. Meshes are simple: either rectangles or circles. Material options include bread pudding, steel, and concrete. The use selects the material type, mesh, and initial position/velocity of the object.
Once the two objects are set on a collision course, the symmetric contact algorithm (implemented in HPX) ensures the conservation of momentum and the mesh surface is deformed accordingly.
Symmetric contact algorithms rely on dynamic, adaptive, and irregular computations. These algorithms have non uniform data access patterns which prevent the applications from taking advantage of data locality and fully utilizing the parallelism offered by the underlying system. HPX provides data access asynchrony without blocking the critical execution path. It does this by using futures based data access thereby achieving the on-demand access needed to more fully exploit the underlying resources in symmetric contact algorithms.