1st SCIEnce Workshop

From SCIEnce

Jump to: navigation, search

The first Workshop of our European Project SCIEnce was held from Monday January 8 to Wednesday 10, 2007. It took place in École polytechnique, located in Palaiseau, 20 km south-west from Paris and was hosted by the Computer Science Lab.

All groups of the project were represented at the workshop, as well as all associates. Apart from presentations by members of the project, we had 2 short communications and 2 invited lectures by externals:

Gene Cooperman home page Disk-Based Parallel Computing: A New Paradigm

Paulo Mora de Freitas and Igor Semeniouk (Nuclear and High Energy Particles Lab of Ecole polytechnique): presentation of the Grid project ¨GRIF¨ web site, part of the european project EGEE web site





                Demo and talk on MapleNet and Web Services. 
                Presentation of activities at RISC-Linz.  Slides (PDF).
               Interacting with remote services, Authorisation,
               Security, Safety.  Slides (PDF).
               Presentation on the annual reporting and NA1 Project. Management meeting.  Slides (PDF)


               Disk-Based Parallel Computing:  A New Paradigm. Slides
               Kevin Hammond: JRA Overall/Status Report (PPT) 
                Dana Petcu: Grid Services versus Web Services (PDF)
              Georgiana Maciaru: Grid Services Example (PPT) 
                Microsoft "High Performance Computing Institute"  
              "Black Forest Grid" 
               TeXmacs and Mathemagix. Web site


                Presentation of GRIF.  Slides (PPT). 
                Abyd Al Zain: Report on GridGUM/SymGrid-Par (PPT) 
                Near-Term Planning for JRA

Annual Reports and Administration

Amongst other things, Steve gave a presentation on the upcoming annual report.

Use of Wiki

Bernd gave a presentation on the Wiki and the Web site.

JRA Workshop and GRID Presentations

We held a workshop describing progress on SymGrid and describing parallel/Grid technologies that could be of interest to symbolic system providers. Gene Cooperman described how parallel disk arrays could be used to simulate large-scale memory, with similar bandwidth to modern RAM (PC-3200). He gave an invited talk on " Disk-Based Parallel Computing: A New Paradigm ". Slides.

Summary: One can observe that 100 local commodity disks of an array have approximately the same streaming bandwidth as a single RAM subsystem. Hence, it is proposed to treat a cluster as if it were a single computer with tens of terabytes of data, with RAM serving as cache for disk. We discuss appropriate data abstractions and algorithms that allow us to hide the high latency of disk and of network. We also briefly summarize other recent activities of our working group: disk-based searches over tens of terabytes of data; progress toward the showing that Rubik's cube can be solved in 20 moves or less (sometimes called ``God's number); a particular a transparent, library-based checkpointing package for multi-threaded, distributed process; lessons from supporting parallelization of the million-line software, Geant4 (from CERN); and support for a task-based parallel abstraction that is both flexible and easy to program.

Researchers from the Nuclear and High Energy Particles Lab of Ecole polytechnique described their Grid application for collecting supercollider information (GRIF -- Grid Ile de France) -- large amounts of data must be stored, but the application is easily parallelised (each independent task takes approx. 30s, there are however 1.2x10^9 tasks per year, each of which produces several MB of data). This might be a possible testbed application for SymGrid-Par in the medium-term, showing that it can be adapted to more traditional scientific settings. Slides.

We were informed by the head of Computer Science that Microsoft provides funding to support HPC research. We may wish to pursue this if we eventually support Microsoft-based Grid systems (see http://www.microsoft.com/emic/nextgrid.mspx for details).

The JRA workshop presentations were as follows:

Kevin Hammond: JRA Status Report (PPT) 
Kevin outlined the structure of the JRA, indicated progress to date, and described the new top-level design of the SymGrid middleware that we are constructing.
Dana Petcu: Grid Services versus Web Services (PDF) 
Dana explained the additional features that Grid Services provide compared with more familiar Web Services, notably a systematic approach to stateful computations.
Georgiana Maciaru: Grid Services Example (PPT) 
Georgiana introduced her prototype implementation of the Science Grid Services framework. This is written in Java, and may either be linked against or called through a shell interface. It is highly generic.
Abyd Al Zain: Report on GridGUM/SymGrid-Par (PPT) 
Abyd showed performance results from GRIDGUM, demonstrating that GRIDGUM could deal with heteroegenity in processor speed/memory across a geographically wide area (Germany and Scotland). He also showed that it was possible to achieve real absolute speedup (albeit only of a factor of 2.5 on 5 processors on the test case) over sequential GAP computations.

On Wednesday afternoon, the JRA core team met to work on the design of SymGrid, and to consider the content of a paper based on this design for submission to EuroPar 2007.

NA3 Discussion

Jürgen gave a presentation on MapleNet and Web Services.

Steve's slides to start the NA3 discussion.

Project Papers

The JRA core team is writing a project paper describing the objectives of the JRA for submission to TFP 2007. Once complete, this will be included on the project web site, and may be abstracted for descriptive information.

One concrete result of the workshop is that there is now a new joint publication among 3 members of the project and our invited speaker, Gene Cooperman (to appear in EuroPar-07):

Kevin Hammond, Abdallah Al Zain, Gene Cooperman, Dana Petcu, and Phil Trinder, 
SymGrid: a Framework for Symbolic Computation on the Grid.
Personal tools
SCIEnce Project