Computational steering for high performance computing applications on Blue Gene/Q System

Abstract

The traditional workflow in a high performance computing (HPC) simulation is to prepare the application's input, run the simulation, and visualize the simulation results in a post-processing step. By performing these steps simultaneously, significant development and testing time can be saved. Computational steering provides the capability to direct or re-direct the progress of an HPC application at run-time by modifying application-defined control parameters using a steering client application. In this paper, we discuss a computational steering framework for the Blue Gene/Q system that provides an innovative solution and an easy-to-use platform, which allows user(s) to connect to and interact with running application(s) in real-time from native desktop steering applications and/or mobile devices. This framework uses RealityGrid as the underlying steering library and adds several enhancements to the library to enable steering support for the Blue Gene systems. The Blue Gene supercomputer presents special challenges for remote access because the compute nodes reside on private networks. This paper discusses an implemented solution for remote steering of simulation applications running on a high performance computer system and describes the implementation challenges.