Introduction

One of the main objectives of the SIMUMAT program is to create a Computing Laboratory to perform the numerical simulations and computations to support the group's current research lines. To meet this goal, a cluster has been recently acquired, which has been called "ODISEA".

ODISEA The cluster ODISEA consists of 16 dual nodes in Rack. Each dual board has two 64 bits Intel Xeon EMT64 processors at 3,2 Ghz. This makes a total of 32 processors available.

The master node has two fast Ultra SCSI 320 10000 r.p.m. disks with a storage capacity of 146 Gb each. Each of the slave nodes has a 250 Gb serial ATA disk (which gives a total of more than 1,5 Terabytes of storage capacity). In addition, each computing node has 2 Gb of RAM.

The nodes are connected by a Gigabit network for system tasks and an Infiniband network (with a bandwidth of more than 16 Mb/sec) used to perform the communications in the parallel programs implemented in MPI.

Currently, the cluster is being used by more than 30 users of Universidad Autónoma and Carlos III de Madrid, the Consejo Superior de Investigaciones Científicas (CSIC) and the Instituto Nacional de Técnica Aeroespacial (INTA).

It is planned to extend this cluster in the near future with 8 additional dual nodes with the same type of processor architecture as the current ones, reaching a total of 32 processors that may satisfy the initial necessities of computational infrastructure of the different research groups that take part in SIMUMAT.



Software

The cluster runs LINUX (RedHat 4.0 Enterprise Edition version) operating system.



Queue system

Users can work interactively through the shell or, in the case of jobs that may need a huge amount of resources (CPU time, memory or disk space), they may use a queue system. The work scheduler (queue system) used at ODISEA is PBS (The Portable Batch System).



Compilers

Several compilers are available at ODISEA, including the GNU compilers gcc for C and C++, and g77 for Fortran, both in the 64 bits version that allows to take advantage of the 64 bits architecture of the Xeon EMT64 processors.

It is also possible to perform the parallelization of codes using the Message Passing Paradigm with the MPI libraries, using the MPI implementation from Scali (ScaliMPI), which is tailored to benefit from the bandwidth of the Infiniband Technology. It is also possible to perform shared memory parallelization, at each node, using OpenMP supported by the Intel compilers.

Intel compilers have also been installed, including Intel C and C++ Compilers (icc) and Intel Fortran Compiler (ifort). These compilers are tailored to produce binaries specifically optimized and efficient for the Intel XEON EMT64 architecture, and also provide implementations of the OpenMP Standard that allows the programmer to make shared memory implementations. Also, the Intel Fortran Compiler fully supports the Fortran 95 language standard, as well as the previous Fortran 90 and Fortran 77 standards, so it is possible for the users of ODISEA to program using any Fortran Standard.


Numerical libraries

Several open-source numeric libraries are installed, for example: - Python with the libraries PythonMPI, Python Numerics and Python f2Py, that make possible the implementation of efficient Python programs, using MPI and rewriting the most time consuming tasks in Fortran modules.

Statistics library R and its parallel version Rmpi, to perform statistical computations and visualization.
Additional numeric libraries will be installed in the near future.
The Matlab program for simulation and scientific visualization, is also available.

Licenses for commercial software such as Fluent, the statistic library IMSL, etc. will be purchased with funds provided by SIMUMAT.