The use case “OpenFOAM” will leverage MIKELANGELO’s virtualization stack to flexibly compute aerodynamic maps for aircraft designs in the cloud. This use case validates MIKELANGELO’s technology stack at PIPISTREL, a Slovenian aircraft manufacturer. PIPISTREL designs aircrafts, which essentially requires computational fluid dynamics, which in turn is very compute intensive. PIPISTREL uses OpenFOAM, which is a popular open source software for computational fluid dynamics, to analyse all of its new aircraft designs. In aeronautical engineering, OpenFOAM allows to compute aerodynamic maps for new airplane designs. An aerodynamic map calculates the lift, drag, and moment characteristics of an aircraft with respect to the incoming airflow at various attitudes. In our context, OpenFOAM takes a basic mesh, changes simulation parameters in order to achieve various combinations of incoming flow angles, and computes the flow of air, resulting forces, and moments for each simulation case. Currently, PIPISTREL typically runs all cases serially, each one on few tens of cores. To cover all the possible aerodynamic attitudes, and to obtain a good aerodynamic map, a few hundred cases need to be run. Accumulating over all cases, the complete simulations require significant computational resources, as each case requires hours or even days to compute, depending on the problem. However, SMEs such as PIPISTREL do not have large amounts of computational resources. Thus, PIPISTREL currently either runs many consecutive cases on a local machine or on a remote cluster. In either case the target machines need to be specifically configured to run OpenFOAM requests. Especially on remote clusters, the costly configuration, often poses as a significant entry barrier.
In addition to a flexible deployment on remote machines, managing large number of cases poses a challenge. Considering the amount of individual cases, we require a tailored management tool, to manage the simulation process. The complete process requires us to prepare all cases, initiate their calculation, monitor for convergence of algorithms, monitor for failures, and, finally, to compile the results from all cases.
Despite the effort, it is important to compute aerodynamic maps in the early phases of aircraft design. A quick and interactive evaluation of an aerodynamic map uncovers controllability and performance flaws, early in the design process. Uncovering flaws early on allows for a detailed analysis of flaws and therefore a rapid design iteration process. Controllability flaws often are very expensive to rectify when they are uncovered in a mature aircraft design, such as a prototype. Controllability issues that have been discovered in a late design phase have been the cause for cancellation of aircraft design projects in the past.
In this use case, we will use MIKELANGELO’s stack to quickly deploy small OpenFOAM cases on remote machines. We will build on the lowest layer with sKVM, leveraging its high I/O performance to transfer case data. Our goal is to provide a complete OSv port of OpenFoam application, fully integrated with the advance I/O features of the MIKELANGLEO stack especially inter-VM communication over shared memory mechanisms to provide further boost of I/O performance. We will use the OpenStack integration and the along with the HPC port of the batch system to OSv to manage the OpenFOAM cases. As part of this integration, we will develop a management system, which will deploy and monitor the computation of cases. At the end of each case runs, the management system will compile the information from all computed cases. By distributing the cases across many machines, we will quickly obtain and compile information about aerodynamic maps. These results, in turn, will give us valuable insights to prepare the following series cases. Thus, this use case will help to speed up the design process of new aircraft. The improved speed in design iterations will help us to improve the quality of PIPISTREL’s aircraft, by detecting flaws earlier on. In addition, the management system for OpenFoam will reduce the management overhead for OpenFOAM simulations. As a consequence, the responsible aeronautical engineers will be able to focus better on actual aircraft design, instead of managing HPC applications. Conclusively, aeronautical designers, especially in small and medium companies with small HPC budgets, will be able to improve the quality of their designs and build better aircraft.