Grants STARE
On Feburay 27th, the NWO Physical Sciences Council granted four projects in the research programme STARE. This programme stimulates innovative multidisciplinary research on the borderline of computer science and astronomy.
The granted projects are:
‘ASTROVIS: Scalable analysis and visualization of high-dimensional astronomical data sets’, Thijs van der Hulst (RUG) and Jos Roerdink (RUG)
‘ASTROSTREAM: Stream Processing of Astronomical Data’, Henk Sips (TUD) and Kjeld van der Schaaf (ASTRON)
‘SCARIe: Software Correlator Architecture Research and Implementation for e-VLBI’, Huib Jan van Langevelde (JIVE) and Cees de Laat (UvA)
‘D2G2: Dutch Dynamic GRAPE Grid’, Simon Portegies Zwart (UvA) and Peter Sloot (UvA)
A more detailed description of these projects can be found below:
‘ASTROVIS: Scalable analysis and visualization of high-dimensional astronomical data sets’, Thijs van der Hulst (RUG) and Jos Roerdink (RUG)
Data sets in astronomy are growing to enormous sizes. Modern surveys provide not only image data but catalogues of millions of objects, each object with hundreds of associated parameters. As an example, in 2006 the OmegaCAM wide-field camera will routinely produce a million multi-dimensional data base entries per night. To explore these data sets effectively, new tools must be developed that can cope with the sheer data volume which has entered the tens of terabytes regime. To maintain applicability, these tools need to be scalable.
Specifically in this research project we will develop feature extraction and visualisation techniques for high-dimensional data. The goal is first to use multiscale operators and then extend these multiscale operators to multi-parameter data, so they can help localise patterns in the high-dimensional astronomical data at all scales and in all directions. Additional goal is to develop interactive visualisation of highdimensional parameter spaces, recognising the power of the human brain.These tools will be tested and qualified on a few small pilot projects. The Sloan 4th data release will be ingested at OmegaCEN and matched to a 70 deg2 WSRT radio survey of the CVn regions providing a 30,000 galaxy multi-dimensional catalogue. Secondly, the tools will be applied to the RAVE survey, to the Geneva-Copenhagen catalogue of nearby F and G stars and to high-resolution N-body simulations of the formation of galaxy halos. Next, the tools will be applied and further developed on very large datasets, such as provided by the sExtractor [3] catalogues of the OmegaCAM instrument, which will contain multi-parameter data for 100 million of detected galaxies.At the University of Groningen astronomy research groups, computer science research groups and a strong astronomical software development team are already collaborating and form the ideal breeding ground for the proposed research. The close interaction between the different disciplines is essential for success. The project will also strengthen the existing collaboration enormously.
‘ASTROSTREAM: Stream Processing of Astronomical Data’, Henk Sips (TUD) and Kjeld van der Schaaf (ASTRON)
The ASTROSTREAM project aims at creating an application development environment for stream-based astronomy applications tailored to the needs of the next-generation radio telescopes LOFAR and SKA. The ASTROSTREAM software architecture will be optimized for high productivity, high efficiency, multi-platform support, and flexibility in run-time behavior. ASTROSTREAM is intended for challenging applications with many Tera-bits/s of streaming input data and tens of Teraflops of processing power, typically on large cluster computers.The ASTROSTREAM framework consists of four parts: (i) on-line stream programming, (ii) grid-based stream coordination, (iii) off-line streaming, and (iv) demonstrators. The on-line streaming is based on a SP (series-parallel) parallel programming model with stream-based and event handling extensions. ASTROSTREAM will allow existing transformational code to be embedded in an XML framework. The stream handling parallel programs will be coupled to a symbolic performance estimator that is used to provide user feed back and will be used for automatic load balancing.The grid-based coordination layer deals with methods to express global coordination of distributed parallel streambased applications. Off-line stream processing will investigate self-calibration techniques for multiple data streams stored on disks. The A
STROSTREAM framework will be validated with two demonstrators containing model-in-theloop astronomy applications.‘SCARIe: Software Correlator Architecture Research and Implementation for e-VLBI’, Huib Jan van Langevelde (JIVE) and Cees de Laat (UvA)
Very Long Baseline Interferometry (VLBI) has undergone a remarkable development. Previously, to obtain the maximum bandwidth possible, data were recorded at telescopes onto magnetic tapes and shipped to the Joint Institute for VLBI in Europe (JIVE) in Dwingeloo. There the data were played back and processed by the European VLBI Network (EVN) correlator, a custom-made supercomputer developed and operated by JIVE. Over the past few years this method has been largely replaced by the use of PCbased hard disk arrays. The easy access to digital data provided by disk technology, combined with the ever-increasing capacity of internet connections, opened the possibility to use the internet to directly connect the telescopes to the correlator at JIVE (so-called e-VLBI).
The many advantages of e-VLBI, scienti c and logistical, led to a proof-of-concept (PoC) project involving JIVE, various radio-astronomical observatories, NRENs (National Research and Educational Networks) and GEANT (the pan-European research network operated by Dante). The PoC succesfully demonstrated the feasibility of connecting several telescopes across Europe in real-time to the correlator at JIVE, albeit at rather modest data rates.
The next step forward is currently underway in the form of the EC funded I3 EXPReS project, which aims at transforming the EVN into a fully functional global e-VLBI network. SCARI will complement and strengthen this ongoing development, concentrating knowledge and know-how locally while taking advantage of extensive international collaborations.
The aim, a robust, real-time e-VLBI network based on distributed software correlation on the NL Grid, will not just involve the engineering of an astronomical instrument, but will keep Dutch astronomers and computer scientists on the forefront of ICT developments in e-VLBI.
‘D2G2: Dutch Dynamic GRAPE Grid’, Simon Portegies Zwart (UvA) and Peter Sloot (UvA)
We plan to explore the efficiency of Grid-based hybrid wide area computing in data-intensive astrophysics problems which are performance limited. For this purpose, we plan to construct a hybrid distributed supercomputer which operates a dedicated virtual organization facilitated via a Grid middelware and smart resource brokers. The backbone infrastructure will be based on the DAS-3 Grid computer, which is a next generation distributed computer across 5 sites in the Netherlands. Several of the DAS-3 nodes (2 to 4 per site) will be equipped with special purpose GRAPE-6Af hardware to realize a heterogeneous setup, allow for asymmetric configurations and obtain the raw supercomputer power required for our astrophysical simulations.
With this facility we plan to study high-performance and high throughput computing, and to explore the efficiency of a distributed asymmetric and hybrid computer with dedicated components.
We are particularly interested in the possibilities of computing on demand by allowing smart broker middleware to allocate and deallocate resources dynamically, during runtime of a simulation. In a practical application, like simulating the dynamics of Galactic nuclei, this allows a much more effective use of the available resources without loss of efficiency.