Grid-computing in mutidisciplinary CFD optimization problems
The challenge of multi-physics industrial
applications
Toan Nguyen
INRIA
Projet OPALE
655, Avenue de l’Europe
Montbonnot
F-38334 Saint Ismier
Phone : +33 4 76 61 52 40
E-mail : Toan.Nguyen@inrialpes.fr
Click here to view the
PowerPoint presentation
Key-words : grid-computing, parallelism, simulation,
cluster-computing, CFD, multidiscipline optimisation.
The advent
of grid-computing technology fosters essential breakthroughs in current problem
solving environments. While compute and data intensive applications
requirements put severe demands on current computing systems, it is clear that
the future of affordable computing resources lies in networked clusters of
smaller, cheaper and cost saving computers. They have altogether the power of
highly specialized mainframes for only a fraction of their cost. Examples of
such systems, currently connecting thousands of powerful but smaller computers
abound. They are supporting highly demanding applications, in the fields of
simulation for powerplant systems, meteorological,
biomedical and climate modelling, as well as complex engineering in the
aerospace and automotive industries.
So far,
these simulation applications were at best formed by complex sequences of
simulation codes in particular domains of expertise. A nuclear plant would be
modelled by a structural mechanics simulator, loosely coupled with thermal,
fluid mechanics, chemical and electromagnetic simulators. This would entail long-lasting
simulation runs on powerful mainframes.
While
adequate parallelisation of part of these simulator codes improves performance
significantly, tight
coupling remains an open issue in such simulations due to the large number of
variables implied in the various simulation models and the implied model
updates it requires.
Besides,
while innovative methods have proved relevant to simulation and optimization,
e.g. evolutionary methods such as genetic algorithms and game theory, their
impact is still limited by the huge amount of computing power required by some
more traditional fields, e.g. mesh adjustment algorithms, although these also
benefit from advances like domain decomposition.
In parallel
with these technological breakthroughs concerning the methods and mathematical
models used in simulation and optimization, several quantum leaps have also
occurred in the last decade concerning the computer science arena.
The most
recent are grid-computing and cluster-computing environments. It has been well
publicized that they allow CPU intensive applications to run in fractions of
the time required previously by super-computers formed by specialized hardware
and software. And indeed, large production simulation environments include
today tens of PC-clusters, each one in turn made of thousands of casual PCs,
all connected by high-speed gigabits/sec networks.
This will
undoubtedly change the future of simulation in the industry, because even SME
are now potential clients of this affordable technology.
This talk
is mainly oriented towards the cross-fertilization between the applied maths,
simulation and optimization communities on the one hand, and the computer
science community, on the other hand, in particular the grid-computing
community.
In the first
part, we put some recent advances in the computer science field in perspective,
relative to the new era opened by PC-clusters, high-speed networks and
grid-computing. We emphasize on the benefits of this technology in simulation
and optimization applications, when parallel programming techniques are used to
implement the core application codes.
In the
second part, we detail some applications related to CFD optimization that can
benefit from the grid-computing technology. We stress the need for an adequate
use of innovative technologies to design and implement new methodologies in the
area of CFD optimization. Examples such as multidisciplinary multi-airfoil
optimization are mentioned to illustrate the challenges of current industrial needs : for example the optimization of multi-criteria
(i.e., high lift and low-drag) devices made of leading edge slats, main airfoil
element, and trailing edge flaps in landing and cruise configurations, coupled
to noise reduction criteria in aircraft design.
We also
emphasize on the user requirements, stressing the need for simple interfaces
that alleviate the idiosyncracies of distributed
computing to the application experts. It is clear indeed that defining, configuring,
deploying, monitoring and maintaining distributed applications is not the
expertise of the application engineers, therefore the interfaces should be
automated and user-friendly to their best.
In the
third part, we describe some potential multidiscipline optimization
applications that require this technology to be practical in the industry,
focusing on demanding industry needs for large scale, real-time and responsive
dynamic simulation environments, e.g. digital aircraft mock-ups. Their impact
on the organization of concurrent, hierarchically organized and overlapping
projects is also mentioned, stressing the need for sophisticated project
management approaches and tools, e.g. virtual organizations.
In
conclusion we draw future research lines for the grid-computing technology and
the multidiscipline optimization fields that promise to enlarge significantly
their impact in complex engineering applications, such as aerospace design.
Invited lecture,