Research collaborations

Research visits

Member Organization
Laura Prada Camacho Computer Engineering Group, Texas A&M University
Abstract
The investigation carried out during the research visit is centered on power management across heterogeneous devices by using automatic generation of policies.

 

 

Member Organization
Francisco Javier García Blas HLRS
Abstract
The objective of this work is the design and implementation of a new file system independent collective I/O optimization based on file views: view-based collective I/O. View-based collective I/O has been implemented and evaluated inside ROMIO im- plementation of MPI-IO standard. The evaluation section shows that view-based I/O outperforms the original two-phase collective I/O from ROMIO in most of the cases for two well-known parallel I/O benchmarks. This is especially due to a smaller cost of scatter/gather operations, a reduction of the metadata overhead, and a smaller number of collective communication and synchronization primitives used in the implementation.

 

 

Member Organization
Francisco Javier García Blas Argonne National Laboratory – Mathematics and Computer Science Division
Abstract
The objective of this work is the design and implementation of an asynchronous data-staging strategy for file accesses based on ROMIO, the most popular MPI-IO distribution, and ZeptoOS, an open source operating system solution for Blue Gene systems. We described and evaluated a two-level file write-back implementation and a one-level prefetching solution. The experimental results demonstrate that both solutions achieve high performance through a high degree of overlap between computation, communication, and file I/O.

 

 

Member Organization
Soledad Escolar Díaz Sensor Networks and Pervasive Computing. Universidad de Bonn (Alemania).
Abstract
The main research field of the Sensor Networks and Pervasive Computing group is the research area of Cooperating Objects. Cooperating Objects define the combination of embedded systems, wireless sensor networks and pervasive computing technology into coherent systems. The main goal is to establish a solid technical and theoretical foundation for systems that combine the traditional strengths of the three different system types.

 

 

Member Organization
Alberto Núñez Covarrubias HLRS
Abstract
EC-funded invitation for visit within the project HPC-Europa from July 15, 2008 to October 15, 2008. The visit is funded by the HPC-Europa project, supported by the European Commission. The purpose of the visit is to carry out work on the project entitled “New strategies for Characterizing and Improving High Performance I/O Architectures” at the host department “Applications, Model Tools” of Alexander Schulz, for which High-Performance Computing (HPC) resources at HLRS would be used. (University of Stuttgart, Germany)

 

 

Member Organization
Alejandro Calderón Mateos HLRS
Abstract
Working together HLRS people (Rainer Keller and Alexander Schulz) and UC3M people (Alejandro Calderón -me- and Alberto Alelu), we designed, developed and benchmarked a PVFS2 prototype with fault tolerant support based on the model that I proposed on my PhD thesis.

 

 

Member Organization
José Daniel Garcia Università degli Studi di Modena
Abstract
Not Available

 

 

Member Organization
Florin Isaila The University of Chicago
Abstract
Not Available

 

 

Member Organization
Javier Fernandez Muñoz HPC-Europa
Abstract
Not Available

 

 

Member Organization
Alberto Núñez Covarrubias IBM
Abstract
Internship at IBM Almaden Research Center, Storage department, from June 22, 2009 to September 17, 2009. San Jose, California, United States.

 

 

Member Organization
Francisco Javier García Blas Argonne natinal Laboratory
Abstract
 

 

 

Member Organization
Luis Miguel Sanchez Garcia HPC Europa
Abstract
In the last years several parallel file systems (PFS) have been developed to offer high I/O performance services to the applications. A PFS distributes a disk partition over a network, allowing a file to be striped over several I/O nodes. This approach increases the performance and scalability of the system. However, existing parallel file systems (PVFS[1], Lustre[4], Expand[3], GPFS[2]) for clusters are still relatively problematic in two basic requirements the users are interested in: cluster-wide coordination and fault-tolerance policies. We propose in this paper a new architecture that combines the advantages of the parallel file systems with the advantages of the traditional I/O architecture of a large cluster, existing I/O solutions for a cluster-wide global management by the means of an I/O scheduler.

 

 

Member Organization
Borja Bergua Guerra LIG Laboratory, INRIA
Abstract
Determine accuracy of simulation network models for the Amazon Cloud. Make adjustment or new platform/network models as necessary. Characterization of tasks length in MapReduce applications.

 

 

Member Organization
Luis Miguel Sanchez Garcia HPC Europa
Abstract
The purpose of this project is to perform an evaluation of these parallel programming models using several C++ benchmarks for exploiting the features of modern computers. We will analyze the behavior of both C++ libraries on NUMA architectures, where data locality is a relevant issue concerning performance. In this project, we will compare these technologies and we will combine both to improve the performance on these computer architectures.

 

 

Member Organization
Juan M Tirado INRIA
Abstract
 

 

 

Member Organization
José Daniel García Sánchez University of Texas A&M
Abstract
Funded by Fundación CajaMadrid worked on refactoring C/C++ applications for parallelism and the impact on performance.