from each other. An example is a Monte Carlo simulation in which many
repeated stochastic evaluations are distributed to several machines and
returned to a single machine for collating and analysis. Any resource analysis problem involving multiple independent simulations can be conducted
in this manner, with the main constraints being the control of the distribution of tasks to various machines and the load balancing required so that
the final compilation of results is not delayed by machines that are slower
than others. This method is appropriate to problems like the evaluation of
multiple alternative scenarios.
Grid computing is somewhat more complex, involving not just simultaneous use of processing power, but the heterogeneity of resources available
across a grid of machines (Foster and Kesselman 1999). An example would
be the activation of and downloading of real-time data from a remote
sensor, the automated processing of a query to a database for related
data located on one machine, providing all of the assembled data as input
to a simulation on a second machine, and processing the output of the
simulation for visualization and analysis on a third machine. The major
challenge in grid computing is the development of a software interface
(middleware) to allow a user to analyze a problem without having to know
the details of where the software, databases, available central processing
unit (CPU) cycles, and other resources are located on the grid. The ideal
system would allow a resource manager to pose a question (with appropriate constraints) and the middleware to assign appropriate components
to different machines on the grid, automatically handling load balancing,
error checking, collating, and returning of the results to the user. For
example, a question might be posed regarding the effects of different landuse patterns in the future of water demands in a region. The middleware
would request land-use history maps from a GIS database, send these to a
machine for spatial analysis, and conduct a simulation to project alternative
futures [as is done in the LUCAS system; see Hazen and Berry (1997)]. The
middleware would concurrently obtain information on water-use history
from a different database, correlate this information with land-use patterns,
combine the water-use and land-use simulations, and provide the results to
the user. Such middleware is well beyond current capabilities, but the software technology needed is developing rapidly (see the GLOBUS project
at http://www.globus.org).
Alternatively, parallelization methods speed processing by breaking
the problem into pieces that can be processed separately. Many ecological
modeling problems clearly fit within this framework, including problems
involving repeated simulations with alternative inputs, sensitivity analyses
obtained by varying simulation parameters, and uncertainty analyses
obtained by including or excluding certain model components or assumptions. Another benefit of parallel architectures is an improved ability to
model situations that are essentially parallel in reality. Ecological systems
are inherently parallel because many components vary concurrently in time
8. Evolving Approaches and Technologies
149
repeated stochastic evaluations are distributed to several machines and
returned to a single machine for collating and analysis. Any resource analysis problem involving multiple independent simulations can be conducted
in this manner, with the main constraints being the control of the distribution of tasks to various machines and the load balancing required so that
the final compilation of results is not delayed by machines that are slower
than others. This method is appropriate to problems like the evaluation of
multiple alternative scenarios.
Grid computing is somewhat more complex, involving not just simultaneous use of processing power, but the heterogeneity of resources available
across a grid of machines (Foster and Kesselman 1999). An example would
be the activation of and downloading of real-time data from a remote
sensor, the automated processing of a query to a database for related
data located on one machine, providing all of the assembled data as input
to a simulation on a second machine, and processing the output of the
simulation for visualization and analysis on a third machine. The major
challenge in grid computing is the development of a software interface
(middleware) to allow a user to analyze a problem without having to know
the details of where the software, databases, available central processing
unit (CPU) cycles, and other resources are located on the grid. The ideal
system would allow a resource manager to pose a question (with appropriate constraints) and the middleware to assign appropriate components
to different machines on the grid, automatically handling load balancing,
error checking, collating, and returning of the results to the user. For
example, a question might be posed regarding the effects of different landuse patterns in the future of water demands in a region. The middleware
would request land-use history maps from a GIS database, send these to a
machine for spatial analysis, and conduct a simulation to project alternative
futures [as is done in the LUCAS system; see Hazen and Berry (1997)]. The
middleware would concurrently obtain information on water-use history
from a different database, correlate this information with land-use patterns,
combine the water-use and land-use simulations, and provide the results to
the user. Such middleware is well beyond current capabilities, but the software technology needed is developing rapidly (see the GLOBUS project
at http://www.globus.org).
Alternatively, parallelization methods speed processing by breaking
the problem into pieces that can be processed separately. Many ecological
modeling problems clearly fit within this framework, including problems
involving repeated simulations with alternative inputs, sensitivity analyses
obtained by varying simulation parameters, and uncertainty analyses
obtained by including or excluding certain model components or assumptions. Another benefit of parallel architectures is an improved ability to
model situations that are essentially parallel in reality. Ecological systems
are inherently parallel because many components vary concurrently in time
8. Evolving Approaches and Technologies
149