Cluster Design and Computational Resources
As CFD capabilities and dependence grow so does the demand for high performance computing resources. Optimum Aero has designed and employed what we believe to be the best value cluster design available to meet our computational demands. Our cluster design utilizes inexpensive, off-the-shelf components that are found in high end gaming desktops and workstations. This hardware is installed in 3U sized cases and rack mounted to minimize footprint. Individual computers, or nodes, are linked together with an Infiniband network fabric to minimize latency and offer linear performance scaling. Rack mounted battery backup units are essential to preserve work during a power outage and to ensure confidence to “set and forget” large run queues.
While this solution may not be ideal for the computational demands of large corporations it is perfect for small to medium sized businesses, consulting firms, or university research labs. Utilization of this cluster design as well as NASA developed CFD software allows Optimum Aero to offer excellent value to our customers with no sacrifice in solution quality.
It has been a steep learning curve to reach the ease of operation and usability we currently enjoy. We are interested in saving others time so they can focus on their design projects and not worry about the hardware, contact us if you would like to request a quote for a custom designed and built turn-key cluster. We have had a number of clusters running with the same hardware for over 3 years. Over this time we have replaced one node’s power supply — an excellent maintenance record.
An example of a recent cluster build with the following specifications is given below. For a half airplane (symmetrical) solution, we are able to run a full drag polar overnight for a medium size mesh. See the image below for a screen shot of the cluster dashboard.
- Computer Specifications for each node — (12 nodes and one master node of the same build)
- 6 Core i7 processor (78 total cores)
- 32 gigabytes of DDR3 1600MHz RAM (408 GB total)
- 3U case for stackability
- Uninterruptable power supple (UPS) sized to support desired number of nodes
- 24 Port Mellanox Technologies Infiniband Switch
- Infiniband Cards
- Software
-
- Rocks 6.1 – Open Source Linux Cluster Distribution
- NASA FUN3D
Rocks Cluster Monitoring Tool – All Nodes Ramping Up
CFD Grid Generation and Adaptation
Optimum Aero specializes in unstructured grid (aka mesh) generation as well as troubleshooting difficult and tedious geometries. Creating advanced meshes throughout our career has afforded us numerous tips, tricks, and workarounds to properly discretize the control volume using industry standard best practices. Grid generation is performed using NASA’s GridTool software combined with years of experience meshing internal and external aircraft geometry, helicopter componentry, and wind turbine systems.
More recently we have been utilizing a procedure known as adjoint-based grid adaptation for certain flow solutions. This process starts with a baseline grid and evaluates the sensitivity of each grid node location to an engineering variable of interest such as lift or drag coefficient. For areas which show high sensitivity to control volume discritization, more nodes are added to minimize the error associated with lack of grid resolution. This is especially useful for high lift configurations with flaps or slats or drag prediction where shock waves must be resolved. See the links below for applicable and interesting data:
- Adjoint-Based Adaptation Applied to the Drag Prediction Workshop II Wing-Body contributed by Beth Lee-Rausch, NASA Langley
- Adjoint-Based Adaptation Applied to High-Lift Airfoil contributed by David Venditti and David Darmofal, MIT
NACA 0012 Grid Adaptation Test Case
Geometry Cleanup
Imported CAD geometry is often unsuited for CFD or FEM meshing. This is a common complaint from many CFD users. Often the CFD analyst must return the geometry back to the design engineer and request that modifications be made to align surface edges, repair holes, trim periodic surfaces, amongst others. The issue is that often times the designer is unaware of the end goal that the CFD analyst requires. The goal is commonly described as a “water-tight” mesh within the industry and is rarely presented on the first iteration.
At Optimum Aero, this is not a new challenge and we have solved the problem by learning to perform the geometry cleanup ourselves. Letting the CFD analyst alter the geometry cleanup has two main benefits and one drawback. The benefits are that the analyst has a clear picture of what is required to readily build a water-tight mesh and also he can make these modifications while ensuring that the surface contours are not altered in a way that changes the external geometry. The drawback is that this time consuming process takes time away from more important and engaging tasks. Optimum Aero understands meshing and can expedite the geometry cleanup process for various applications. See the louver example below which shows CAD imported IGES surfaces, cleaned up geometry, computational mesh, and finally a pressure contour plot used for load determination.
Geometry Cleanup Process: CAD Import – Surface Cleanup – Mesh Generation – Solution