 |
|
DAILY NEWS AND INFORMATION FOR THE GLOBAL GRID COMMUNITY /
|
Applications:
GRID SPEEDS COMPUTATION ON MATLAB MATHEMATICAL SOFTWARE
Dai Nippon Printing Co Ltd in conjunction with Shin Seiichi, associate
professor of Department of Information Physics and Computing, Graduate School
of Information Science and Technology at The University of Tokyo; and Katsumi
Konishi, lecturer at The Center for Continuing Professional Development of
Kogakuin University, have developed a system for high-speed computation of
mathematical computing systems used in university research facilities.
The newly developed system facilitates the high-speed computation of MATLAB
algorithms. MATLAB is a high-level technical computing language and
interactive environment for algorithm development, data visualization, data
analysis and access, and numerical computation. MATLAB is developed and
commercialized by The MathWorks, the world's leading developer of technical
computing software for engineers and scientists in industry, government and
education. MATLAB is used today by more than 1 million scientists and
engineers worldwide, 100,000 of them in Japan.
The DNP developed Grid computing system AD-POWERs, which distributes and
decentralizes MATLAB operations over multiple PCs connected to the network,
and repeating the aggregation of the computing results carried out by the
various PCs.
AD-POWERs was used in the decentralized processing and distribution area,
along with the result aggregation portion of the tasks allotted to the PCs
participating in decentralized processing. AD-POWERs is a software for Windows
PCs capable of high speed processing, by automatically detecting PCs in
offices and research facilities that have become idle as their user steps away
from his or her desk, leaving the computer with the screen saver running, and
dispersing the processing load to those PCs for parallel processing. Priced at
an affordable 209,790 yen per network segment, AD-POWERs is capable of
accommodating up to 253 PCs in a single network segment for parallel
processing. This newly developed system is mounted with a function for calling
up MATLAB functions for use with AD-POWERs.Using this new system, MATLAB and
AD-POWERs are pre-loaded into PCs participating in decentralized processing.
And the user can carry out parallel processing merely by programming with
MATLAB.
The new system was evaluated at Tokyo University and Kogakuin University using
a grouped vehicle control problem involving multiple vehicles, and aiming to
find the minimum-sum of the vehicular routes for all vehicles provided with
different start and finish points on a specific route, which also included
multiple no-entry points. It took 180 seconds to compute this problem using a
single PC, but only 28 seconds with the new system using eight PCs, reducing
processing time to almost a sixth of the original.
As can be observed from in the case of grouped vehicle control problem
involving multiple vehicles, with simulation designed to elicit an optimum
from among a variety of possibilities, it is necessary to repeat such tasks as
carrying out of a large number of computations and comparisons between them;
something that can take up to three days in the case of complicated processes.
That may also require start-up costs and running costs running into millions
or even hundreds of millions of yen if super-computers or specialized
decentralized processing systems are employed to do the job. And as it is also
necessary to write the program in a specialized language, it is not possible
to apply such assets as existing programs.
As the newly developed system uses computers that have fallen idle there is no
need to invest in new hardware, and as it is possible to switch to a
decentralized programming environment using multiple computers from individual
processing over a single PC with just minor alterations to programs, the new
system is also expected to contribute to the speeding up of research
activities.We also anticipate major needs as the way is now open to run MATLAB
applications with the gigantic memory areas that it was difficult to run on
single PCs given their OS and CPU limitations, and possible to carry out
computing based on detailed and high-precision models, while also aiming for
increased computing precision and economies of scale. The fruits of this
development effort are likely to be seen as DNP plans to undertake marketing
activities, starting with client systems, and finally moving to offer a MATLAB
compatible AD-POWERs package. In the first instance we aim to offer a
solutions package to the automobile industry, which sees heavy use of MATLAB.
|