Development and implementation of a high performance computer (HPC) cluster for engineering education simulations

Abstract

With the advancements in high performance computer (HPC) computing, it is only natural that engineering education also utilizes the massive computational capabilities of large server clusters to enhance student learning. This paper presents recent work in developing and implementing complex engineering simulations for engineering education. Key aspects of this work include developing methods to access the simulations through web pages, creating user friendly input modules (web-based), automated job control system for web submission, efficient core utilization for a large number of simultaneous users, and display of results on the user's web page. Each of these issues are critical to engineering education due to the unique environment required for using computers in classroom and lab settings. A detailed working example for torsional stress of non-circular bars is given in the paper to illustrate the implementation of a server cluster. Currently, up to 40,000 degree of freedom problems can be solved with the system. Execution time varies depending on the number of cores devoted to a given problem. But even if only one core is used, the solution time is 10-50 faster on the cluster than on the client device (laptop, smart phone, tablet, etc.) since the cluster solver is compiled C code instead of less efficient Flash ActionScript. All examples used in the paper are currently available at www.eCourses.ou.edu. The paper addresses the special needs of engineering education when utilizing HPC systems. All simulations are web-based, and students do not need special knowledge of clusters, job control, or parallel programming. Simulations are accessed through a web page where parameters, such as boundary conditions, geometry constraints, loads, accuracy and grid resolution (FEA) are specified. The web interface is one of the more difficult aspects of the system. The interface needs to be intuitive and accessible on a large number of devices, such as laptops, smart phones, and tablets. To simplify the development of the user interface, this system used web-enabled Flash Player for both the simulation set up and viewing of the results. This allows most devices connected to the internet to access the system through a common web page. Utilizing Flash also makes it easier to develop advanced user interface graphics such as real-time grid generation, sliders, input boxes and graphical result output. The paper provides details on how the dedicated 32 node (384-core) engineering education cluster was set up using Windows 2008 HPC Server R2. This includes the job control system, allocation of core resources, cluster solvers, utilization of math libraries, and network communications between the cluster and user during the solution steps. One of the primary goals of this paper is encourage others to pursue developing complex simulations for engineering education so that students can reap the benefits of the recent advances in cluster computing. © 2012 American Society for Engineering Education.