 |
 |
|
| |
|
|
|
|
|
|
OverviewFEARLUS Model 0-6-4 is an agent-based model of land-use change developed at the Macaulay Institute in Aberdeen. The system contains objects that represent human decision-makers in the real-world (land managers), and takes into account attributes such as yield from land parcels. Parameters to the modelling environment allow a variety of land-use strategies and their outcomes to be explored. In this ESRC funded pilot demonstrator project we have explored the costs and benefits of using proposed Semantic Grid standards and methodology; these focus on supporting high-level e-science activities rather than low-level computational operations and promise to allow scientists to work with “information and knowledge” rather than “raw data”. Central to this vision is a need for shared representations which facilitate communication of experiment specifications, parameters and results. Our work has included:
We have developed FEARLUS-G – a collection of services to perform land use simulation in the Grid context based on the existing FEARLUS land use model and utilising Globus Toolkit 3. We have also developed a Web client application (My Workspace, see Figure 1) that makes use of the services by uploading and running FEARLUS simulations; the services are also accessible through an extended version of the existing FEARLUS desktop client. 
Figure 1. The My Workspace Interface. We chose to deploy FEARLUS as a Grid service for a number of reasons: firstly, the Grid infrastructure allows large simulation experiments to be distributed across the Grid to make use of unused processing power; in fact, a typical FEARLUS experiment already involves running a series of simulations, but on a single machine. Secondly, the use of meta-data combined with Grid infrastructure allows us to create a co-laboratory where land-use scientists and others interested in rural policy can access the FEARLUS-G service and share and reuse results and observations. Figure 2 presents an overview of the FEARLUS-G system architecture, which is built on top of OGSA 3.2.1 (Globus Toolkit 3) and Web services . 
Figure 2. The FEARLUS-G System Architecture. The FEARLUS-G system is composed of two main components: The Web/Grid Services component and the FEARLUS Model Interface component. The Web/Grid Services component contains the core services that manage access to the computational and data resources. These services are as follows:
The Fearlus Model Interface component contains the interface between the Web/Grid services and the existing FEARLUS agent-based simulation software. This interface serves a dual purpose: firstly, it converts the meta-data from the Grid services into low-level simulation parameters and uses those parameters to perform a simulation run with the FEARLUS agent-based simulation software; secondly, it converts the numerical results from the simulation back into meta-data to be stored in the Repository Service. The meta-data managed in FEARLUS-G is supported by an ontology which captures the concepts and relationships important to scientists performing their research activities. The ontology includes a collection of generic elements that are intended to be applicable to any e-science application, and also elements that are specific to simulation modelling and FEARLUS-G in particular. Further details of this ontology and the relevant supporting technologies can be found in the Recent Publications section and Tools & Links section. We have conducted an evaluation of FEARLUS-G with the help of a group of "assessors" from the international scientific communities interested in agent-based social simulation, land-use and water management modelling. During the first phase of the evaluation (meta-data evaluation) assessors were asked to use the FEARLUS-G interface to create and apply meta-data to Grid based land-use simulation experiments. The results of this initial evaluation activity are available in the Tools & Links section. The next evaluation phase (usability evaluation) aims to test the wider usability of the FEARLUS-G Grid service, in a research capacity (uploading FEARLUS models, comparing results from assessors’ own models with those from FEARLUS, checking results in FEARLUS publications, or exploring FEARLUS's capabilities and limitations as a stage in the process of designing a new model), or as a teaching tool. Tutorial material and an associated questionnaire are available in the Tools & Links section. Through our experiences to date we have learnt valuable lessons about the provision of Semantic Grid tools to the computational land-use modelling community: Our Grid service provides scientists interested in land-use phenomena with a means to run much larger-scale experiments than previously possible on standalone PCs, and also gives them a Web-based environment in which to share simulation results. The Grid is clearly a valuable tool for managing the computation involved in running large-scale experiments. In fact the Globus environment is ideal for simulation modelling because the cost of a simulation run is less that the cost of the communication necessary to run a distributed experiment and aggregate the results. However, the Globus environment is not well-suited for the fine grained exploratory work that the existing FEARLUS desktop application is able to achieve. Scientists still want to use the patterns of workflow related to the desktop application. For this reason, we have extended the existing desktop application to be able to upload models to the Grid service, creating a link between the exploratory tool and the Grid experiment platform. We have defined an initial collection of ontology elements that describe the scientific objects necessary to enable collaboration between members of a community of land-use scientists. Our ontology represents generic scientific concepts such as hypotheses and experiments, as well as more domain-specific concepts tailored to the use of FEARLUS. Social scientists are able to run FEARLUS experiments on the Grid and collect the results; a private interface (My Workspace) allows them to manage their activities (experiments, investigations, etc.) while their work is in progress. Scientific resources can also be made public, allowing others to investigate them further. Not all categories of users require access to the same meta-data. For example, we have exposed FEARLUS model parameters to land-use scientists familiar with the existing FEARLUS application in such a way that the complexity of the Grid services is hidden to them. For this reason we provide, together with the desktop application, the My Workspace interface which allows users to access and manipulate low-level data. However, we also need to provide high-level data to promote accessibility to the wider scientific community. For this reason we also provide access to our Public Repository where resources contain high-level generic properties (e.g. describedIn, agreesWithHypothesis, hasDescription). Throughout this project we have encouraged computational land-use modellers at the Macaulay Institute to lead the process of engineering the FEARLUS ontology. This important activity forced them to think about what conducting an experiment actually meant. Prior to this project commencing, PERL scripts were used to conduct experiments with FEARLUS and the experimental design was thus hidden. One positive side-effect of the ontology building exercise is that the experimental workflow is now exposed, making its use transparent to other users. Development of the ontology has also encouraged them to begin thinking more generally about meta-data to support agent-based social simulation. |
NEWS04/07/2005 22/06/2005 10/06/2005 13/05/2005 24/03/2005 31/08/2004 - 03/09/2004 23/07/2004 08/04/2004 14/01/2004 |
