Due to the growing importance of simulation
and the availability of appropriate software in industry, the use of
parallel systems has now become standard in the commercial environment
as well. Large parallel computer systems are used by financial service
providers, in the oil industry, and in Formula 1 racing. Most systems
today consist of standard PC hardware, in combination with a high-speed network.
The ITWM belongs to the pioneers concerning the application of PC clusters for
industrial simulation problems. It is currently operating parallel systems with
more than 1,000 CPUs for the development of parallel software and the solution
of industrially relevant computation tasks.
Technological problems have stopped the continuous increase in performance
by faster clock rates; dual core, quad core, and more are the current key
words for a further increase. The parallelization of software has thus become
a key technology in all IT relevant areas. Future efficient parallel software
will have a hybrid structure, on the one hand making optimum use of multicore
subsystems, and on the other hand scaling over many computers.
The new Cell Processor, which has been produced in cooperation by the
companies IBM, Toshiba, and Sony and is working in the new PlayStation,
is a prototype of this new generation of CPUs. In a strategic cooperation
with IBM′s development laboratory in Böblingen, the CC HPC develops applications
and analyzes the Cell platform with respect to its suitability as a HPC system.
The IBM Cell Cluster, which has been installed at the ITWM in June 2008, made
it into the Top500 with 1,120 cores, and even comes first place on the list
of the most energy efficient systems (www.green500.org).
The extension of research activities in the field of multicore systems was
one of the main subjects in the year 2008. The project activities of the
CC HPC are currently divided into four fields:
Parallelization and performance analysis - seismic
The focus of activities in 2008 lay on the cooperation with our partners
IBM and StatoilHydro. By combining fundamental HPC know-how and practical
knowledge, the CC HPC has succeeded in developing particularly high-performance
codes that meet great interest among clients. The Competence Center was especially
successful in 2008 in the area of seismic, with a long-term cooperation agreed
upon with the Norwegian group StatoilHydro.
Grid computing - cloud computing
In the year 2008, grid computing turned to cloud computing. New services
such as the Amazon Elastic Compute Cloud also offered new possibilities for
us to apply our technology. Our individual grid solution PHASTGrid was ported
to the Amazon Cloud and presented as a productive platform during trade fairs
and talks already very early. We are dealing with the license management of
cloud and grids on the basis of an innovative idea, currently developing a
respective product.
The CC HPC is an active participant in D-Grid and the European EGEE project.
One focus of activities is the further development of basic grid technologies
and the participation in the grid standardization boards. Within a second
main subject, Fraunhofer grid solutions have been further developed in
cooperation with the Fraunhofer partners within the internal enterprise
grid project. We have been able to apply PHASTGrid within different cooperation
projects with industrial partners. A productive environment for the assessment
of the quality and performance of the grid infrastructure has been made available
to the D-Grid community by the development of the grid benchmark portal
jawari.itwm.fraunhofer.de.
Visualization
In the field of CPU-based ray tracing methods, commercial products are meanwhile
available. With PreStack-Pro, we were able to present an innovative software
product for the quantitative analysis and visualization of large prestacks in oil
industry at international trade fairs (SEG, EAGE). An entirely CPU-based postprocessor
for the analysis of computation results of the software MAGMASOFT was developed
for the company MAGMA Gießereitechnologie, and delivered to customers.
HPC tools
High-performance tools for the development of HPC software and systems are GraPA,
a parallelization framework for hybrid applications in the area of multicore
distributed memory, FVM, a software interface providing global shared memory on
Infiniband clusters for parallel applications, and the parallel Fraunhofer file
system FhGFS.
The activities with respect to the development of new tools for future multicore
processors have been combined within the Multicore Innovation Center, funded by
the Federal Ministry of Education and Research, the Fraunhofer-Gesellschaft, IBM,
and further industrial partners.
|