Research Projects
KOJAK
KOJAK is a performance-analysis tool for parallel applications
supporting the programming models OpenMP, MPI, SHMEM and combinations thereof.
Its functionality addresses the entire analysis process including instrumentation,
post-processing of performance data, and result presentation. It is based on
the idea of automatically searching event traces of parallel applications for
execution patterns indicating inefficient behavior. The patterns are classified
by category and their significance is quantified for every program phase and
system resource involved. The results are made available to the user in a flexible
graphical user interface, where they can be investigated on varying levels
of granularity. KOJAK is jointly developed by Forschungszentrum Jülich
and the University of Tennessee.
For more information: http://www.fz-juelich.de/zam/kojak/
Part of the KOJAK toolkit is a source-to-source translation tool
called OPARI that is responsible for the instrumentation of OpenMP pragmas
or directives. OPARI can also be used by third-party tools to provide instrumentation
based on the POMP profiling interface for a large variety of measurement tasks.
For more information: http://www.fz-juelich.de/zam/kojak/opari/
Automatic Overheads Profiler for OpenMP programs
Parallel programs often do not obtain ideal speed up. This
is due to time being lost in various overheads, such as load imbalance, synchronisation
costs, remote memory access and replicated code. Ovaltine is a tool to provide
users, automatically, determine such overheads in an OpenMP code. This helps
users to tune their code to a realistic best implementation. Currently, Ovaltine
only supports Fortran 77 but we hope to expand it to also cover Fortran 90 and
C.
For more information http://www.cs.man.ac.uk/cnc/mscprojects/index_2003.html#cnc.p4/
NANOS: Effective Integration of
Fine-grain Parallelism Exploitation and Multiprogramming
The NANOS project at CEPBA (European Center for Parallelism
of Barcelona, UPC) aims at performing research on software technology with the
objective of enabling the efficient use of the different types of compute servers
that will be available in the near future, focusing on the problem of performance
portability. The following architectures are considered: multi-threaded processors,
multiprocessors on a chip and SMPs, high-end NUMA multiprocessors and clustered
SMPs. The research platform under development includes an advanced OpenMP compiler
(NanosCompiler) and runtime support through a set of components (NthLib, the
user-level threads library for multilevel hierarchical parallelism exploitation).
The concept of reusable dynamically computed schedules will be key to enable
dynamic load balancing and improvement of memory locality (through user-level
dynamic page migration). The project is also analyzing the feasibility of going
in the direction of automatically self-adapting programs (which includes low
overhead monitoring, automatic application structure identification and prediction).
Finally, application-kernel cooperation mechanisms, which include dynamically
measurements of application efficiency, and the combined space/time-sharing
scheduling policies aim at offering an efficient environment for the execution
of multiprogrammed workloads.
The NANOS project is being funded by the CIRI (CEPBA-IBM Research
Institute) and the European Comission (through the POP Performance Portability
of OpenMP IST project, project partners: CEPBA/UPC, Istituto di Cibernetica/CNR,
INRIA and LHPCA/Univ. of Patras).
For more information http://www.cepba.upc.es/nanos
The ParaMount Group at Purdue University
The ParaMount group conducts research projects related to
High Performance Computing. The current reseach projects are:
The Polaris Parallelizing Compiler
P-Grid
ADAPT: Automated De-Coupled Adaptive Program Transformation
The Moerae Portable Interface
The Ursa Minor / Ursa Major Toolset
Cepheus - Parallelization of non-Fortran languages
Performance Forecaster (PerFore)
Comprehensive Characterization of Industrial Application Performance
Benchmarking with large-scope applications
Compiling for Speculative Architectures
For more information http://cobweb.ecn.purdue.edu/ParaMount/
INTONE: Innovative OpenMP Tools
for Non-Experts
The main objectives of the INTONE project are: 1) To define
an instrumentation interface for OpenMP compilation and execution systems. 2)
To develop a powerful performance analysis system for OpenMP applications, consisting
of a graphical performance analysis tool, an advanced Performance Assistant
assisting the non-expert user, and the necessary instrumentation facilities.
Emphasis will be on ease-of-use for non-specialized users. 3) To advance the
state-of-the-art in programming shared-memory clusters by extending the OpenMP
API by performance-oriented communication statements creating a Cluster Programming
System integrated with the performance analysis system. 4) To evaluate these
components with application software from industrial application writers, making
sure that end-users needs are met. From this evaluation, it is planned to demonstrate
that OpenMP, the Cluster Programming System and the performance tools can considerably
improve the efficiency of writing, tuning and maintaining parallel applications.
And 5), to exploit the results both commercially and through academic channels,
and to disseminate them in the OpenMP community to influence emerging OpenMP
extensions.
The INTONE project is funded by
the European Comission under the program IST. Project Partners: Pallas (project
coordinator, technology integrator and exploiter), CEPBA/UPC, KTH, TU-Dresden
( OpenMP technology providers), Enel.Hydro and LMS International (industrial
application developers).
For more information http://www.cepba.upc.es/intone
The High Performance Computing Tools group
in the Department of Computer Science
at the University of Houston is relatively young. It started when Dr. Barbara
Chapman joined the department in the Spring of 1999. Since then we have grown
into a team of almost twenty people, working on projects in the following areas:
For more information http://www2.cs.uh.edu/~hpctools/
Please, provide us with information about your OpenMP project.
| 
