Complex, large-scale, collaborative simulations are becoming more and more crucial for decision making in critical situations like floods, earthquakes, wildfires, terroristic attacks, epidemics, pandemics, instabilities in financial markets and similar. At the same time recent advances in experimental techniques such as detectors, sensors, and scanners have opened up new vistas for real-time initial data collection and aggregation in many levels of details. It allow to study the connections and interrelations between various systems of the real world taking into account their social and socio-technical impact in frame of complex network (CN) models.
The principal feature of CN modeling in relation to critical phenomena is the strong sensitivity to some scaling effects (e.g. for reliable evaluation of critical transitions in N-element system, the 100N-element system must be used for numerical simulation). Thus, it requires to use supercomputing technologies, but their direct application is restricted by connectivity of CN due to various social links (as limitation for parallel execution). This effect is especially visible for critical transitions which are mainly depend on a lot of far links in CN. Therefore, not only the algorithms and software, but also the simulation model itself must be developed for the efficient simulation of critical phenomena in large social networks.
• The general multiscale model for social CN simulation is proposed on the combination of three coupled models which are:
• Approximate analytical model of social CN on the macro-level, which allows to perform a qualitative assessment of availability of critical conditions.
• Rough direct model of CN at the micro-level, based on a model with a relatively simple topology reproducing an algorithm that allows to study the basic properties of the social system in a steady state.
Detailed direct model of CN at the micro-level, which takes into account heterogeneity and hierarchical features of social system, necessary to describe the dynamics of critical states.
The main advantage of using three models together is that the approximate model allows us to estimate the macro-characteristics for individual fragments of the original network, which makes it possible to identify the most volatile fragments (with a view to further detail), as well as fragments possessing relative independence. Each of them can be further reproduced or rough or detailed model of CNs on the micro-level, depending on the probability of the occurrence of a critical state. Thus, the use of an approximate model allows not only to decompose the parallel algorithm for direct models, but also to create rules for scheduling and load balancing, which in turn ensures the efficiency of using the resources of a supercomputer or even distributed environment (Grids, clouds etc.). It allows to implement this approach to early warning systems in frame of Urgent Computing (UC) paradigm – a new area of computer science addressing algorithms, methods and tools enabling prioritized, immediate and effective access to large compute and storage systems for such emergency computations which require clever decision making. UC is being considered as computational services (or resources) since data services work together in distributed computational environment to help decision makers create an optimal behavior scenario within a strict time limit. The Virtual Laboratory for Urgent Computing (VLUC) is developed for support the UC-computations in various domain areas, including social and socio-technical systems. The applications to multichannel spreading of infections, rumor spreading through social networks in Internet, and evolution of the “dark” networks (for example, criminal networks of drug distribution) will be presented.
Acknowledgements. This research is partly financially supported by Government of Russian Federation, Grant 074-U01 and Project 11.G34.31.0019 (220 Decree).