Applications

GriddLeS is currently being applied to a range of applications, including Atmospheric Sciences and Computational Mechanics.

Atmospheric Sciences

Figure 1 shows a Grid workflow application that performs atmospheric modelling. This sample application takes temperature and pressure data from a variety of instruments, such as satellites and airborne and seaborne sensors, and feeds these to a range of different numerical models. In particular, data is assimilated into a general circulation model of the atmosphere (1), which computes the flow fields across the entire globe. This global model in turn drives the boundaries of a regional weather model (2) which produces more accurate wind vectors and temperature and pressure fields over a limited area. These values are in turn streamed into a variety of pollution models, such as a photo-chemical pollution model (3), a particle dispersion model (4) and a bush fire model (5). Each application addresses some particular aspect of the atmosphere in isolation, but when linked together they interact and provide a rich set of data ranging from weather to pollution. For example, a bush fire generates particles that must be dispersed, and also increases various precursors that affect photo-chemical pollution. If the fire is severe enough, it actually affects the regional weather. Accordingly, the different models need to interchange data at various times.

In the Grid, static data sources, such as pollution inventories and vegetation maps, required by the various computational models might be distributed geographically, but copies may be available at more than one site. This means that when models are scheduled to the various machines in the Grid, the location of the closest data also needs to be taken into account.

Figure 1 - Atmosheric Sciences grid Workflow

Computational Mechanics

This application considers computer models of thin plates containing holes and subject to cyclical loading. The models assume pre-existing cracks normal to the hole profile and use the Jones method of crack dynamics to estimate the number of cycles required for these cracks to spread from an initial length to some final length. Our aim is to determine the hole shapes that will maximize the life of the worst (least cycles) crack. Previous work has shown that optimizing for life in this way may give different results from optimizing for stress on the hole boundary. The picture on the right shows the stress distribution in the plate for a particular hole shape.

In order to complete the computations, we need to execute a pipeline of 5 programs, as shown on the right. CHAMMY takes a formula for a hole shape, depending on several parameters and generates points on the boundary of that hole. The programs MAKES_SF_FILES and OBJECTIVE are used to transform data from one phase to the other. PAFEC is a finite element code that computes the stress tensors in the meshed design. FAST is a crack propagation code that computes the number of cycles before a number of independently placed cracks reach a certain length.

Traditionally, the entire pipeline has been executed on the one computer, with intermediate results passed using files. Importantly, some files are passed from one phase to another, whereas, other files are simply read from the file system. The final output, RESULT.DAT, contains the value for the life of the design, which is the minimum time for any of the cracks to reach a certain length. This result defines the life of the design. The GriddLeS File Multiplexer and GNS are flexible enough to map some files to the local file system, whilst linking writer-reader file chains into direct socket connections.