Environmental Cloud Compute


When thinking about how modelling and simulation can help the environmental and sustainability agenda, it's sensible to ask how much does doing the modelling and simulation contribute to the challenges it's aiming to aid? This is particularly relevant when thinking about which cloud computing provider to use.


The picture of power consumption and carbon dioxide emissions of the three biggest players in cloud computing provision is a tangled web to say the least, and it's important to understand their wider service provision outside of compute and HPC for computer aided engineering.  In terms of environmental impact there's three factors at play: the efficiency of the servers; the efficiency of the infrastructure (especially the cooling system); and the source of the electricity.


All of Google, Amazon and Microsoft have pledged to de-carbonize their data centres, but all still utilize fossil fuels to provide the majority of their power. The mechanism for the de-carbonization pledge is the use of renewable energy credits (RECs), which are sold by renewable energy companies as a measure of the societal value of the renewable energy, in addition to the sale of the energy itself - essentially a subsidy to the renewables industry. RECs allow Google to match all of their electricity consumption since 2017 with these "renewables purchases".


Google have also been addressing the cooling infrastructure costs by using machine learning to optimize it's data centres with an algorithm that has learnt to adjust the cooling system in response to the local environment, sampling the weather every five minutes in the process. Microsoft's approach to infrastructure has been a little more adventurous - making a sealed shipping container sized data centre and lowering it in to the ocean in their Natick research project. This removes the need to use fresh water to cool the servers, using heat exchange technology from submarines that benefits from the low temperature variability of the sea. It also allows the data centres to be located in close proximity to offshore wind turbine farms.


The ability to locate data centres close to renewable energy sources is a significant advantage for smaller cloud providers with a lower server footprint. A few examples stand out: H66 have sites around the globe, with the claim that all use local hydro-power; Green Mountain (see what they've done there?) locate servers in Norwegian mountains, use local hydro-power and provide efficient cooling with local Fjord's; atnorth use renewable sources at their locations in Sweden and Iceland. Whether the compute services offered match up from a cost and usability perspective to those of the big three is another question, but from an environmental perspective it's certainly worth taking a closer look.

Falling in at the other end of the neat software names list to last months entry, is the very aptly named solids4Foam. As you might guess, it's a solid mechanics library for OpenFOAM. Solid mechanics capability has never been a prominent feature of the OpenFOAM toolset, despite the solidDisplacementFoam solver being featured in the seminal FOAM paper. The culprit is the use of the finite volume method (FVM) by OpenFOAM for solving partial differential equations, as opposed to the finite element method (FEM), which has historically been the approach of choice for solids mechanics solvers.


The finite element method has extended it's use in to fluid dynamics problems (the non-linear convective term in the Navier-Stokes equations makes it a little more involved) with commercial "multi-physics" FEM solvers such as Comsol and Hyperworks. There has been little reciprocity from the finite volume world making headway in to the solid mechanics domain, until solids4FOAM.


From a physics perspective solids4FOAM is very capable, offering a wide range of elasticity models, fracture and contact modelling as well as fluid structure interaction (FSI) capabilities. Each of the solid mechanics and fluid dynamics solvers are written as C++ classes, allowing the selection of the solver, or combination of solvers, of choice in the input dictionaries.


One really neat capability is the adaptability to use across all three main forks of OpenFOAM: the foundation version, the ESI version, and the extend version. Whilst written in the extend framework, it will compile under both other versions and make some required adjustments at run-time. solids4FOAM is available at a bitbucket repository and you can read more about the capability and validation models in this arXiv paper.

Altair acquires Flow Simulator from GE Aviation

Flow Simulator is an integrated flow, thermal and combustion simulation package, originally developed by GE Aviation and, prior to this announcement, commercially distributed by Altair. Flow Simulator uses a one dimensional network modelling approach to provide systems level design and optimization covering applications such as gas turbines, internal combustion engines and processor cooling. It has some neat features, such as a point-cloud recognition system that automates the fitting of the one dimensional network to three dimensional CAD layouts and assemblies and variable model resolution. There's a neat video of the capabilities on the Altair website.


A belief that the ongoing development and support is better placed in the hands of Altair, given their string position in the engineering simulation sector, seems well founded, and likely to enable broader capability enhancements and commericial viability to customers other than GE. It works nicely for Altiair as Flow Simulator has been filling a gap in their portfolio since 2018 under the previous commercial distribution arrangement. It also looks to work nicely for GE Aviation who have secured access to Altair's complete suite of simulation tools under a memorandum of understanding signed as part of the acquisition deal.