<img height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=1875728222659097&amp;ev=PageView&amp;noscript=1">

Stacks On Stacks: Parallel Runs On Single Compute Nodes

Home » Blog » Stacks On Stacks: Parallel Runs On Single Compute Nodes

As computer power becomes more affordable and more accessible, powerful computational fluid dynamics (CFD) techniques are becoming widely used by engineers. 

Their goal is to optimize performance and design, identifying ways to effectively meet deadlines, reduce time to market, and work within budget constraints. A proficient way to optimize performance and design, is to use 'parametric CFD analysis' and experimental testing. In this article, we explore the topic further and assess how EXN/Aero's stacking ability is providing engineers with the option of 'doing more for less'.

Generally, parametric CFD analysis is made up of five key steps including;

  • Problem definition
  • Dimensional reduction
  • Design of experiment
  • Management of CFD simulations
  • Metadata analysis (this may include visualization, optimization, and sensitivity analysis

A recent article provided a good example of where parametric CFD analysis would prove useful within the aerodynamic design sphere, discussing the placement of vortex generators (VG) on the wing of a commercial airliner. VGs are generally used to correct flight attributes that are considered undesirable i.e. low-speed instability. In this example, even a minor change to the configuration could have a significant impact on the aircraft's handling, so complex analysis is required. By using parametric analysis, engineers have a more solid understanding of how a device behaves so can ensure a product is better designed.

In the first instance, engineers must define the problem and identify requirements for the component, system, or process. In the example of an airliner, engineers must ensure their designs allow the aircraft to fly safely and efficiently, using CFD simulation, wind-tunnel tests, and flight tests to achieve this goal. In this example, designers must identify where best to place VGs, an aerodynamic compromise to delay the onset of boundary-layer separation on the wing surfaces (a common problem with commercial jets). While placing a VG in one place may have a beneficial effect in terms of boundary-layer separation, it may create problems in a different area of the flight envelope, potentially altering flight attributes, and overall aircraft performance. Because of this, engineers must make their evaluations throughout the whole flight envelope, and parametric analysis is a key asset to reaching a decision.

As is typical with engineering design problems, the problem definition often results in a large number of dimensions - too many for a reasonable parametric analysis. To reduce this number, many engineers will refer back to previous studies, narrowing down their designs and the number of dimensions to be tested.

In this context, “experimental design” refers to the set of cases that the CFD code will run. The overall goal here is to determine which input factors have a significant impact on performance measures. Another important use of experimental design or design of experiment, is to develop a metamodel (a simplified model of the simulation model) based on the important factors to predict the model response for factor-level combinations that were not actually simulated due to execution-time or setup-time constraints, or because a prediction is needed in real time. A metamodel can also be used to find the factor combination that optimizes the simulation response.

Even by reducing dimensions in the problem definition phase, a small parametric study still results in hundreds of runs. High-resolution CFD solutions have previously been associated with huge costs, particularly those of full aircraft configurations. This had placed pressure on ensuring the problem definition stage was accurate. This is not just the case for aviation. In fact, this is also the case in several other industries where engineers are required to run a large parameter set of simulations including HVAC, automotive, and tidal energy.

EXN/Aero Stacking

To alleviate the aforementioned problems, Envenio has recently introduced an ability for engineers to run multiple simulations on single compute nodes (for the same hourly price). Below is an example of a steady state simulation (external flow of a wingbody) running on a 2 GPU node with 48GB of ram. On our systems, you can fit roughly 1M control volumes per GB of ram (this decreases with additional equations and double precision). In the left graph, you see that you're able to run up to 24 - 2M cell meshes at the same time. Looking over to the right graph you have a total simulation cost of only $19, and a per simulation cost of less than $1. EXN/Aero also features XML scripting for running simulation batches, making it easier to run these low cost parameter sets. 

2017-09-12 13_55_00-Copy of benchmark 2 - Excel.png

*Note: stacking up to 7 parallel runs is available in a standard EXN/Aero on-demand plan, going beyond 7 parallel runs is available upon request. 

Since this solution is on-demand and pay-as-you-go, you can load up these nodes on the fly to alleviate simulation demands. Most people think of EXN/Aero as only being useful for larger and more challenging simulations, but as you can see, there is significant value in using this platform for smaller sized meshes. 


Many of these compute nodes can be activated on the fly, meaning you can tackle a whole parameter set concurrently. In a recent example, a customer was running 20 simulations sequentially for their client. Each 8M cell simulation took roughly 4 hours, resulting in a total delivery time of 80 hours. On our platform, they were able to complete all 20 simulations on only 4 compute nodes in just 5 hours for 20% less cost. This enables consultants or design engineers to iterate quicker on their product or system designs and provides them with access to more valuable engineering information. 

2017-10-04 17_01_22-Benchmark Tool - Excel-1.png

Parametric CFD analysis combined with experimental design, has been shown to play a crucial role in the engineering design of fluid-dynamic devices. Without it, engineering methods can be costly and time consuming. By the time a project is complete, there could already by a more-efficient way to utilize the device. An understanding of the five stages of parametric CFD analysis gives users a better understanding of how a device behaves over its entire operating envelope and quickly produces better designs.

EXN/Aero: For The Future

EXN/Aero is a general purpose CFD cloud solver committed to embracing the needs of the engineering community. By considering the feedback of industry experts, such as those highlighted by CFD Engine, vendors can work to transform this sector and open up new opportunities and innovation.

The EXN/Aero platform includes a meshing tool, the solver, and a post-processing tool, and plenty of storage for your files. There are a range of on-demand options available to users, helping them to better manage project cash flow. This affordable CFD software is sure to be an asset to companies or CFD freelancers like. We offer free trials of the product, starting with a walkthrough with one of our engineers.

Request Demo

2017-10-5 | Categories: CFD, simulations, HPC