Objectives
The INRIA joined the Gelato
Federation in February 2003. This Federation is
working to develop scalable commodity software to enable researchers to
advance their studies in developing and technology-intensive areas, such
as life sciences and physical sciences. Co-founded by HP and
seven of the world’s leading research institutions, Gelato is an open
source community initiative designed to foster the development and
dissemination of focused computing solutions for researchers and
associated IT staffs working on Linux-based
Intel® Itanium® 2 platforms.
INRIA integrated the Gelato Federation in order to help contributing the
community with new research results, helpful tools and solutions.
Expertise
INRIA has been conducting research activities in parallel computing
software for more than 20 years. Nowadays, it has been concentrating its
efforts on design, implementation and test of parallel software for
clusters and grids. Several high-performance computing architectures
have been developed within INRIA, and succeeding into being
registered in the TOP500 list of
the most powerful computer systems in the world.
Leveraging upon its expertise, INRIA is part of the Parallel File Systems, Cluster
administration tools, runtime for parallel programming and
distributed scheduling focus areas from the Gelato Federation. It
covers the following areas:
- System administration and management
- Distributed memory management
- Parallel file systems
- Performance monitoring and visualization
- Runtime for parallel programming
- Distributed Scheduling
- Resource allocation
- Parallel programming models (C++, CORBA or Java)
- Parallel applications
Resources
The institute contributes an
experimentation cluster built around 104
Intel® Itanium® 2 bi-processor machines at 900 MHz, equipped with 3 GB of
memory and 72 GB of local storage, interconnected using
a Myrinet-2000 network and directly connected to the VTHD network.
- NFSp
is an
attempt to build a parallel File system that implements NFS. It is able
to distribute the data handling load over the multiple machines of a
cluster in a very efficient way.
(Contact:
Yves Denneulin
, project
APACHE
)
- Ka-tools allow the realization of large scientific computing
clusters while being based on software known as "free" (i.e. under GPL
license or comparable). The objective consists of the creation of a
GNU/Linux distribution for clustered machines, meeting the needs for
deploying, administering, and programming clusters for the purpose of
intensive calculation.
(Contact:
Philippe Augerat
, project
APACHE
)
- Athapascan is a high level data-flow language, 100% C and
C++ compatible, with explicit parallelism and enabling various
scheduling strategies.
(Contact:
Jean-Louis Roch
, project
APACHE
)
- Inuktitut is a Portability layer for Athapascan.
(Contact:
Thierry Gautier
, project
APACHE
)
- Padico is a software environment for computational grids; it
is designed for high performance parallel computing, distributed
computing, and software components. It targets code coupling
applications based on the concept of parallel CORBA objects.
(Contact:
Christian Pérez
, project
PARIS
)
- Kerrighed is a Single System Image operating system for
clusters. Kerrighed targets ease of programming, high performance and
high availability. Ease of programming is achieved as Kerrighed supports
both the message passing and the shared memory programming models.
Kerrighed takes benefit of the underlying hardware performance by
providing global management of all cluster resources (processor, memory
and disk).
(Contact:
Christine Morin
, project
PARIS
)
- INFLATE (Infrastructure for Large scale data servers)
aims at using clusters of workstation
resources for data servers instead of "classical" scientific
computing applications. To successfully run data server applications,
large amounts of data (i.e. large numbers of IO operations) need to
be processed.
Also application servers need to provide scalability, high availability
and failover to enterprise applications in a clustered environment.
INFLATE contributes both at the operating system and middleware levels.
(Contact:
Emmanuel Cecchet
, project
SARDES
)
- Proboscis is a modular infrastructure that enables
description, creation and administration of inter-nodes hard disks
accesses. Proboscis can hence be used to emulate a storage area network
(SAN) and to provide a high performance distributed file system for
clusters.
(Contact:
Emmanuel Cecchet
, project
SARDES
)
- C-JDBC allows any Java application using the JDBC standard
database access mechanism with transparent manipulation of a database
distributed onto a cluster of machines. C-JDBC uses full replication of
the database on all the machines in the cluster, supports heterogeneous
environments and offers scalability to a number of machines.
(Contact:
Emmanuel Cecchet
, project
SARDES
)
Contacts
For overall information about INRIA's activities regarding Gelato please
contact
Brigitte.Plateau@inria.fr.
For technical information please
contact
Bruno.Richard@imag.fr.
