Page 230 - Fundamentals of Ocean Renewable Energy Generating Electricity From The Sea

Ocean Modelling for Resource Characterization Chapter | 8 219

to Telemac and ADCIRC—two very popular unstructured models. Blue Kenue
is very flexible and can handle a wide range of data types, including GIS files,
GRIB, x−y scatter data, and time series. However, one of the disadvantages
of Blue Kenue, in comparison to computing environments which incorporate
a programming language like Matlab, is that although you can save all the
variables in the workspace as you proceed, it is not possible to save a ‘line-
by-line’ sequence of commands. Therefore, if one develops a model grid of a
region and wishes to go back several steps in the processing (e.g. to remove an
island from the domain), many procedures must be repeated manually, rather
than editing and running a script. Note that although primarily used to deal with
unstructured (triangular) meshes, Blue Kenue will also generate rectangular
meshes. Blue Kenue also has a very easy to implement presentation quality
animation facility (including flight paths), helping with model visualization.

8.7 SUPERCOMPUTING

In 1965, Intel cofounder Gordon Moore noticed that the number of transistors
per square inch on integrated circuits had doubled every year since their
invention. Moore’s Law predicts that this trend will continue into the foreseeable
future. We are now in a golden age of scientific computing, when efficient
multicore desktop PCs can be purchased at relatively low cost. However, within
the context of ocean modelling, supercomputing refers to running a compu-
tational task simultaneously (i.e. in parallel) on hundreds or even thousands
of processors. Rather than running a job, representing a single computational
domain, in series, the computational domain can be decomposed into several
subdomains, which can be run together in parallel (Fig. 8.16). Although such a
task is possible on a desktop PC for a limited number of processors, it is when
a model that has been optimized for parallel processing is divided into several
hundred tasks that supercomputing really excels.
The simplest way of preparing a computational domain for parallel pro-
cessing is through tiling. However, a ‘tile’ that contains mostly land cells will
represent a considerably lighter computational task than a tile which contains
entirely sea points—a situation that is known as uneven load balancing. There-
fore, many models use load-balancing techniques for domain decomposition,
as demonstrated in Fig. 8.16 for the POLCOMS model. Each of the 10 nodes
(0–9) shown in this example has 10,281–10,366 ‘wet’ grid cells, representing a
departure of no more than 0.5% from the mean–excellent load balancing.
If we were to run a serial task 1000 times, for example, with slightly differing
initial conditions for each simulation, we would complete our simulations in
1/1000 of the time that it would take to complete the same tasks in series—
a considerable time saving, in a situation that is known as ‘embarrassingly
parallel’. This would represent perfect ‘linear’ speedup where
Time for 1 processor
Speedup = (8.45)
Time for n processors