The power of HPC for solving problems is unrivalled, but it is not always easily accessible for small companies and those working in fields with low levels of computing expertise. The company Polyhedra Tech is aiming to change this through a PRACE SHAPE project. Its co-founder Pau Fonseca i Casas explains how they are optimising the SDLPS simulator for use in a number of different fields.
Polyhedra Tech is a company that builds simulation tools to help SMEs optimise their day-to-day operations and increase their performance. This can involve improving energy efficiency, consultancy for the construction industry, sustainability services, near-zero energy buildings analysis, power energy systems, cloud simulation and service analysis, and certification for buildings. The core of their business model is the use of a distributed simulator called SDLPS that enables automatic translation of models defined using SDL into working simulations, which can then be parallelised. SDL, which stands for specification and description language, is an ITU-T standard language intended to be used to represent reactive systems, such as telecommunication or simulation systems. Thanks to its flexibility, SDL allows for quick and accurate translation of real-world contexts into formal modelling frameworks, providing evidence-based solutions to complex problems. It improves the verification and validation of models, therefore increasing the robustness of predictions and solutions. This methodology, currently used by Polyhedra Tech, but originally developed in the InLab FIB research centre at the Universitat Politècnica de Catalunya, has helped a considerable number of the company’s clients, on projects ranging from the design of Terminal 2 at Barcelona el Prat international airport, to improving avalanche predictions.
To connect a model that describes how one system works to another model that describes another system, you need a formal language. SDL is excellent for this.
Pau Fonseca i Casas has been working on a PRACE SHAPE project to help iron out some of the difficulties that Polyhedra Tech is facing when working with SDLPS on supercomputers. “When we have a graphical representation of a model that has been defined using SDL, it is complete and unambiguous,” he explains. “But, when trying to turn this model into code that can be executed on a supercomputer, lots of problems can appear.”
“For example, we need to know what libraries can be used to codify these models. We also need to know how we are going to deal with the code that represents this model, and how we are going to connect the different elements that represent the behaviour of different processors in the system. So despite the fact that you begin with a complete and unambiguous model, there is still a lot of work to do to get it to run on a supercomputer such as MareNostrum. In this SHAPE project, we are trying to solve these problems and define a first conceptualisation of a model that can be used on a supercomputer.”
The project has two main aims. The first is to investigate novel strategies that can leverage an HPC platform such as MareNostrum to accelerate the execution of an SDL graphical model. The second is to test, benchmark and optimise the simulator on social science models requiring different parallelisation paradigms. This will open the use of HPC to a new group of users while at the same time increasing the robustness of their software by improving the level of documentation, verification and validation of the models.
SDLPS is the simulation engine that allows Polyhedra Tech to work with SDL, but it is not able to generate its own code. Fonseca and his team have been working with supercomputing specialists at the Barcelona Supercomputing Centre to create a method that allows it to generate its own code, which they are now testing on MareNostrum.
Figure caption: The methodology used by Polyhedra Tech has been used on projects as diverse as designing airport terminals to predicting avalanches
The models made using this kind of approach are not usually made by computing specialists. For example, Fonseca and his group work closely with people working in areas such as history and geography. When these specialists want to create a model, they use SDLPS to define it. The model defined through this process is a set of graphical diagrams that explain how the model behaves. This model is then codified into something that can be run on a super computer using the libraries generated by specialists at the Barcelona Supercomputing Centre. To validate the codes generated, they will be executed on MareNostrum and compared with outputs executed on normal computers. If they are in agreement, then the codes have been written correctly.
The world is now entering the Fourth Industrial Revolution, characterised by a fusion of technologies that is blurring the lines between the physical, digital and biological spheres, collectively referred to as cyber-physical systems. It is marked by emerging technology breakthroughs in a number of fields, including robotics, artificial intelligence, nanotechnology, quantum computing, biotechnology, the Internet of Things, the Industrial Internet of Things, and more.
One concept that has emerged from this new era is that of the digital twin – a digital replica of a physical entity. In industry, a digital twin refers to a model that is a representation of an industrial process. “A key problem that industry faces with these digital twins is that if you want to combine things in reality, you also need to find a way to connect these digital twins,” explains F onseca. “To connect a model that describes how one system works to another model that describes another system, you need a formal language, and SDL is excellent for this purpose. At lower levels, this can already be seen with the OPC-UA language, which connects devices in the Industrial Internet of Things. But to connect the behaviour of systems, we can use SDL.”
Fonseca and his team have been working on a project called NECADA that uses this approach and SLDPS as a core simulation engine. NECADA takes care of the environmental directives and international rules in the designing process of buildings, optimising the behaviour of building or cities from the point of view of the sustainability. It will require the power of supercomputers due to the number of interconnected systems, and so the work done in this SHAPE project will help it achieve its goals.
The results of the project will improve the usability of SDLPS on supercomputers for end users with different HPC capacities and different levels of HPC expertise available to them, while the testing and integration of social science models will extend the target audience of the simulator. Wider application of SDL provides the potential for streamlining, automating and documenting the process of simulation design and development among non-academics and within academic fields where the level of computational expertise is low. The combination of these two project goals will give interested stakeholders the ability to solve real-world problems by quickly designing and implementing complex social science models and running them using HPC tools without the need for an extensive team of computer science experts.
For more information: www.polyhedra.tech
Resources awarded by PRACE: Pau Fonseca i Casas was awarded 100 000 core hours on MareNostrum hosted by BSC, Spain