Within the Centre of Excellence for Global Systems Science – CoeGSS – specialists in High Performance Computing (HPC) and in Global Systems Science (GSS) are brought together in order to create a new community that can support better decision making about global...
Within the Centre of Excellence for Global Systems Science – CoeGSS – specialists in High Performance Computing (HPC) and in Global Systems Science (GSS) are brought together in order to create a new community that can support better decision making about global challenges. It shall enhance GSS-applications with the capabilities of highly scalable HPC performance.
Global Systems Science is an emerging field of innovation with continuously growing relevance for industries and their decision makers, global politics with their various economies as well as academic sciences and civil society.
To improve accuracy and precision of models, simulations and resulting advice, a higher computational performance is mandatory to reflect the dependencies and to cope with huge amounts of input and output information in reasonable time frames.
So far, High Performance Computing mostly focuses on computationally intensive problems, such as classical computational fluid dynamics, weather forecast, astronomy as well as multi-physics simulations. For this purpose, the systems are shaped into a direction that enables highest performance to run simulations on hundreds of thousands of computational cores to solve the complex problems. However, with the advent of advanced simulation systems based on agent-based models and the resulting complex data correlation, High Performance Data Analytics (HPDA) will be one of the major future challenges. Therefore, the key performance indicators move into a data-centric computation direction: huge amounts of data have to be processed in an efficient, precise and fast manner, which requires assistance of HPC systems.
Starting with the envisioned CoeGSS simulation workflow over the last 18 months, the development on all components has been instantiated. In two rounds of requirements and reactions the needs of three pilot projects have been characterised and adaptation and development of tools has started.
A web-portal has been developed, integrating a number of components: CKAN to provide a data repository, Moodle to organise training, Askbot to manage user support and a Django-based Portal as a way to provide a common frontend. For a comfortable and secure authentication system an LDAP server has been installed.
Parallel to the work on methods, tools and services and in close contact with it, the models for the pilots were developed and refined. With the use of Pandora, an HPC framework for agent-based simulations, the Health Habits and the Green Growth pilot simulations were initially performed and its parameters calibrated. However, at the same time it was clear that Pandora will not be the tool that will be used in the frame of higher scaling simulations, as it shows a variety of drawbacks. Thus the elaboration of alternatives was also performed, trying to find those which may be compliant with Pandora interfaces at the end.
Besides the technical work, the business aspects of the Centre of Excellence have been addressed, too. For this we follow the design thinking approach; one ingredient to sustainability is validation with potential stakeholders. A questionnaire has been designed and first interviews with potential customers have been performed to collect feedback.
With the intention to bring the HPC and GSS community together, the consortium organised an international conference on Synthetic Populations that brought together experts from several data intensive social sciences, like Global Public Health, Urbanisation and Finance.
The project is targeting innovative technologies, and there is a clear progress beyond state of the art for the following domains:
Agent-based modelling frameworks for HPC
Agent-based computer models are becoming increasingly popular in the social sciences, especially as a tool for providing evaluation of planned policy measures and policy recommendations in complex environments.
However, there are just a handful of HPC-ready frameworks and within these the relations between agents so far are mostly comprised to spatial relations. We have designed an architecture for simulations that allow information exchange between agents along their social connections (modelled by multigraphs), leading to major advance beyond the current state of the art.
Synthetic population generation
A synthetic population is a computer representation that statistically matches the actual population, but does not invade anyone\'s privacy. That makes it a valuable input for simulations in disease spreading or travel modelling. Existing synthetic populations mostly have been created for special purposes. CoeGSS advances the state of the art by a general DSL based approach for creating tailored synthetic populations. Resulting in a framework for the creation of synthetic populations that will also include the creation of realistic social networks for the agents, the relevance for extending the state of the art is high.
High Performance Data Analytics
The project advances technologies for data analytics, in particular for highly unstructured and uncorrelated data. This encompasses pre- as well as post-processing steps for the actual application execution, but also novel approaches like on-site data analytics. As a consequence, a high potential for innovation and enhancements of the current state of the art is given.
Social Learning
Many societal challenges like climate change, increasing migration and urbanisation cannot be addressed by application of simple individual means. Instead, complex, manifold and flexible measures will be necessary. Running simulations of different scenarios will allow transparency and participation of non-experts in the decision making process. As the centre’s services will cover the whole workflow from data collection, modelling, and running simulations to immersive visualisation, we are developing a powerful toolbox to support this approach.
More info: http://www.coegss.eu.