Exploitation of High Performance Computing in the FLAME Agent-Based Simulation Framework

The first prototype of the Flexible Large-Scale Agent Modelling Environment (FLAME) has been developed through a collaboration of University of Sheffield and STFC Rutherford Appleton Lab. It is an implementation of communicating X-machines for use in multi-agents simulations. X-agents models can be created to model many types of complex systems and the simulations run on parallel platforms. FLAME is an environment for generating agent-based applications - it is not an agent application in itself. FLAME takes as input from the application development a model definition and a set of agent functions and using the FLAME Template Library, FLAME constructs the agent-based application required. FLAME has been used in a wide variety of applications ranging from the biological sciences to behavioural science and on to logistics and economic modelling. Population sizes range from a few hundred agents to models containing many millions.FLAME has been design to generate application which can run on both serial and parallel computing systems depending on the size of the model and the duration of the simulation. The programs generated by FLAME have been shown to be very portable and versions have been run on a large variety of systems. Parallel versions of applications have run successfully on a range of large parallel systems including HPCx and HECToR using anything from tens to thousands of processors.The major objective of this project is to re-engineer the FLAME system in the light of the authors' experiences in a number of large research project to improve its parallel performance and to improve its functionality and flexibility. The performance gains will be achieved by utilising the multi-threading (multi-core) capability of modern computing nodes and by improving the agent/task scheduling. By re-engineering the FLAME Template library and the model parsing we will be able to introduce new ways of expressing the characteristics of the agents and their interactions.

Agent-based systems are one of the most vibrant and important areas of research and development to have emerged in information technology in the 1990s. Put at its simplest, an agent is an entity that is capable of flexible autonomous action in a dynamic, unpredictable, typically multi-agent domain. An agent could be an individual interacting with others or it could represent an organisation such as a bank. In biological applications agents might be cells or enzymes or even complete organs such as heart or lungs. Many observers therefore believe that agents represent the most important new paradigm for software development since object orientation. The importance over the past ten or so years of agent-based simulation and modelling techniques to both the commercial and academic communities can not be under estimated. Agent-based simulation has permitted a radical shift in which complex systems with many interacting components can be modelled 'from the bottom up' rather than from the 'top down'. A good illustrative example might be something like an ant colony, in which the structure and overall behaviour and capabilities of the community vastly exceed the sum of the simple rules which govern the actions of its individual members. While similar principles can be applied to many ecological populations (flocking of birds would be another good example) the same ideas are also applicable in a tremendous range of scientific applications, for example cells in the human body/ brain, components of an engine (or similar physical example), behaviour of individual decision-makers in an economic market, or interaction between vehicles in a transport network. Through such mechanisms some of the most fundamental challenges in both natural and social sciences are potentially addresses e.g. formation of tumours from individual cancer cells; movement from prosperity to recession through the changing decisions of individual economic actors. The variety of applications using agent-based techniques has grown significantly during this period. The importance of using discrete, interacting intelligent entities to simulate complex non-differential systems has been recognised by a growing variety of disciplines. The world-wide financial crisis and the many natural disasters are two important areas in which agent based modelling will have a significant impact in the future. Major advances in the modelling of social systems have been achieved by Prell (Maryland) who has been using FLAME and HPC on modelling social capital network dynamics. Walkenhaur's group (Rostock) is using FLAME to try to understand the formation of actin filaments on novel surfaces. Slusarek (Stuttgart) has built a very revealing model of ant colony foraging dynamics using FLAME Recently several medical-related projects have started using FLAME: Dynamics of tumours and natural killer cells - Cook (Leeds Medical School); modelling the way tumours control the development of artery growth - Chico; the role of a new famil of proteins, Tribbles, in the immune system - Kiss-Toth; and modelling of the TILLR complex of membrane proteins in the innate immune system - Francis - all at Sheffield Medical Schol. At the latest meeting of the British Ecological Society (September 2010) FLAME was demosntrated and a number of world leading ecologiststs are to start using it - Covich (Georgia) to model shrimps; Colchero (MPI) on jaguars and road traffic; Schoenrogge (CEH Oxford) on ants etc. These are just a few of the scientists using FLAME and its uniqueness in respect to its availbility for HPC is a big attraction. Thsi is an example of where the UK is well in the lead of the rest of the workld.