Section: Other Grants and Activities
HIPerCAL studies a new paradigm (grid substrate) based on confined virtual private execution infrastructure for resource control in grids. In particular, we propose to study and implement new approaches for bandwidth sharing and end to end network quality of service guarantees. The global infrastructure (computers, disks, networks) in partitioned in virtual infrastructures (aggregation of virtual machines coupled with virtual channels) dynamically composed. These virtual infrastructures are multiplexed in time and space, isolated and protected. The goal of this project is to explore an approach in a break with current services-oriented principles developed in grids to jointly enhance the application portability, the communications performance control and their security. The project aims at providing a grid substrate based on end to end bandwidth reservation, control overlay, network and system virtualization, cryptographic identification principles. The proposal is to be validated and evaluated at different scales on the Grid5000 testbed with biomedical applications, demanding in security , performance and reliability. 10 to 1000 processors, links with 100Mb/s to 10Gb/s, few microseconds to 100ms will be involved in these experimentations. We aim at demonstrating the functional transparency, enhanced predictability and efficiency for applications offered by the HIPerNet approach. RESO has developed, deployed and tested the first version of the HIPerNet software on the Grid5000 testbed.
Year funding: 100Keuros
This ANR (Appel Blanc International) started in october 2009 and will end in september 2012. It is a collaborative project between the GIPSA Lab (Grenoble), MOAIS (INRIA Grenoble), RESO (INRIA Grenoble), the University of Osaka (the Cybermedia Center and the Department of Information Networking) and the University of Kyoto (Visualization Laboratory).
It is no falsehood to state that “current society and science attempt to deal with increasing amounts of data". Today, peta-scale data are commonly gathered as well as generated thanks to the continuous development of measurement technologies and computational resources in diverse fields of science and society. Efficient processing or generation of peta-scale data requires high performance computational (HPC) resources which should be made remotely accessible through long-distance high performance networking and might be represented thanks to interactive scientific visualization. Consequently, generation or processing of peta-scale data benefits from the emergence of adequate “Information and communication technologies (ICT)" with respect to high performance “computing-networking-visualization" and their mutual “awareness". In the current proposal, it is aimed to develop and validate such ICT solutions using a transnational high-speed research network between Japan and France connecting GRID5000 (France) to the Naregi (Japan) testbed. Data-transfer protocols are aimed to be validated on data obtained for a real scientific problem involving peta-scale data.
Due to the medical relevance as well as basic scientific interest, peta-scale data are obtained from HPC Computational Fluid Dynamics (CFD) simulations on a vector supercomputer (NEC SCX9 Japan) aiming to predict the airflow through the upper airways. In addition, CFD simulation outcome is used as an input for aero-acoustic computations (CAA) for prediction of noise production. High performance computing is needed in particular to predict fricative noise due to the requested accuracy of the flow field (up to 16kHz). The outcome of CFD and CAA simulations will be validated on flow and noise measurements on a suitable experimental setup (France). Besides the international transfer of the generated peta-scale data, scientific visualization of peta-scale data is aimed on a single PC as well as on a tiled display wall for 3D interactive reconstruction of the flow and noise data.
In summary, the proposed project aims to contribute to the state-of-the-art of HPC, networking, scientific visualization and their mutual interactions for peta-scale data, while at the same time it is aimed to contribute to basic research in the fields of CFD and CAA applied to flow through the upper airways. The current proposal can only be realized thanks to the joint efforts and resources of the French and Japanese partners involved which gathers specialists in networking, middleware, scientific visualization, HPC and upper airway flow modeling and noise production.
Started in october 2008, this ANR project, leaded by J. Barral (Sisyphe, INRIA Roquencourt), is a partnership between INRIA (Sisyphe and Reso), university Paris 12 and university Paris Sud (équipe d'accueil EA 4046 Service de Réanimation Médicale CHU de Bicêtre).
Numerical studies using ideas from statistical physics, large deviations theory and functions analysis have exhibited striking scaling invariance properties for human long-term R-R interval signals. These signals are extracted from electrocardiograms and represent the time intervals between two consecutive heartbeats. The scaling invariance measured on these empirical data are reminiscent of geometric fractal properties verified theoretically by certain mathematical objects (measures or functions), which are called (self-similar) multifractals. These numerical studies also reveal that the scaling invariance may have different forms, according to the fact that the patients have a good health or suffer from certain cardiac diseases. These observations suggest that a good understanding of multifractal properties of cardiac signals might lead to new pertinent tools for diagnosis and surveillance. However, until now, neither satisfactory physiological origin has been associated with these properties nor mathematical objects have been proposed as good models for these signals. It is fundamental for possible medical applications in the future to go beyond the previously mentioned works and achieve a deepened study of the scaling invariance structure of cardiac signals. This requires new robust algorithms for the multifractal signals processing; specifically, it seems relevant to complete the usual statistical approach with a geometric study of the scaling invariance. In addition, it is necessary to apply these tools to a number of data arising from distinct pathologies, in order to start a classification of the different features of the observed scaling invariance, and to relate them to physiological concepts. This should contribute to develop an accurate new flexible multifractal mathematical model whose parameters could be adjusted according to the observed pathology. It is also important to strengthen the information by performing the multifractal analysis of another fundamental signal in cardiology, namely the blood pressure, as well as the simultaneous multifractal analysis/modeling of the couple (R-R,Blood Pressure). This project aims at achieving such a program. It also proposes to contribute to explain the origin of the scaling invariance properties by developing a reduced order dynamical system, which shall describe the heart's electromechanical activity and simultaneously shall generate multifractal outputs in accordance with the R-R signals models. A 1-D model of cardiac fiber would be already very satisfactory. This aspect of the project is closely related to the delicate issue of understanding the link between multifractal phenomena and PDEs, another topic that will be investigated. The project team consists in six members representing four partners: two specialists of multifractal analysis, one specialist of cardio-vascular system modeling and PDEs control, one specialist of statistical signal processing and two physiologists (among which one cardiologist) specialists of cardio-vascular signals processing. The project will benefit of a wide data's bank of long term (24h) R-R interval signals already recorded in various clinical settings including diabetes, acromegaly and sleep apnea, and a prospective data bank will be established in the field of medical intensive care unit, namely in patients presenting cardiovascular pathologies like heart failure, arterial hypertension and chock states. The data bank will include both R-R interval signals and arterial blood pressure signals.
Year funding: 2,5Keuros