Conference Tutorials

The following tutorials will be organized on August 31, Tuesday. The tutorial fee is included in the conference fee. The tutorials will take place in the morning and in the afternoon. The placement of the tutorials during the day is tentative, at the moment.
Registered people will be asked (by email) to subscribe to the tutorials before the conference, just to organize the sessions.

Tutorial 1: Open Source Middleware for the Grid: Distributed Objects and Components in ProActive

Speaker: Denis Caromel (INRIA, CNRS/I3S, Univ. Nice Sophia Antipolis, France) and Romain Quilici (INRIA, CNRS/I3S Univ. Nice Sophia Antipolis, France)
Date: 31 Aug 2004

Dowloads:

Abstract

Object technology and middleware are increasingly used within Grid computing. At the same time, components are becoming a very effective tool for the composition and deployment of business applications. While the difficulties to actually program and deploy on the Grid have been, to a great extend, acknowledged, components technology is in the process of bringing many advances for Grid programming and deployment. At the same time, we believe a strong programming model is still very much needed. The aim of this course is to explain and detail these aspects. The course objectives are: - to explain the main principles of objects and components for the Grid, - to illustrate the concepts and ideas with an effective Open Source LGPL environment for the Grid: ProActive (ObjectWeb consortium) The tutorial will include concrete applications and benchmarks.

Morning:
Introduction:

The ProActive middleware: Conclusion and perspectives:
Afternoon: Hands-on programming session
The afternoon will be a guided hands-on programming session, together with demonstrations. The tutorial attendees should bring a laptop for actually programming and deploying on the Grid. We will secure the access to a few clusters in order to experience on a real GRID! The following real-size application should be used as a concrete use-case: http://www-sop.inria.fr/oasis/ProActive/jem3D.html

  • Parallel and Active Objects
  • Synchronization
  • Group Communications (vs. MPI)
  • Mobility
  • Hierarchical Components
  • Deployment (rsh, ssh, Globus, LSF, PBS, etc.)
  • Security
  • Use case and performances (vs. Fortran MPI) Conclusion and perspectives:
    Afternoon: Hands-on programming session
    The afternoon will be a guided hands-on programming session, together with demonstrations. The tutorial attendees should bring a laptop for actually programming and deploying on the Grid. We will secure the access to a few clusters in order to experience on a real GRID! The following real-size application should be used as a concrete use-case: http://www-sop.inria.fr/oasis/ProActive/jem3D.html

    Requirements for attending the tutorial: in case you want to use your laptop to experiment with ProActive, please come equipped with a 10/100 Ethernet card AND a long Ethernet cable (to connect to the switch) or with a wireless card or, possibly with both.


    Tutorial 2: Achieving Usability and Efficiency in Large-Scale Parallel Computing Systems

    Speaker: Fabrizio Petrini and Kei Davis (Los Alamos National Laboratory, USA)
    Date: 31 Aug 2004

    Downloads:

    Abstract

    Demand for high performance computing continues to accelerate: as greater computational capability becomes achievable new classes of problems become amenable to computational solutions, fueling demand for yet-greater computational power. The remarkable growth in number and size of very large scale computing platforms has been enabled by the ability to construct machines, in a modular fashion and of almost arbitrary size, from readily-available hardware and software components. The design and construction of such machines, generically known as clusters, is in sharp contrast to earlier supercomputers, designed from the ground up, with highly specialized, monolithic architectures. Today the situation is such that it is far easier to build such a machine than it is to make effective use of it; indeed, practical problems such as fault tolerance and effective resource utilization are such that, for example, the U.S. Department of Energy, the largest single consumer of supercomputers, has capped peak capability at 100 tera-ops for future procurements until solutions to these problems are found. In this tutorial we describe the current state of practice for constructing and managing such machines. Drawing from analytical and experimental studies, we describe in some depth the problems--some predictable, some unexpected--that arise at large scale. Significant attention will be paid to current solutions, and to research directions and efforts for better solutions at current scale and effective solutions at larger scale. The treatment will be particular to a range spanning purely commodity clusters, larger machines such as ASCI's Q and Lightning, and more specialized architectures such as BlueGene/L, but also relevant to machines of similar classes.

    Tutorial 3: Grid Resource Management and Scheduling

    Speaker: Ramin Yahyapour (University of Dortmund, Germany)
    Date: 31 Aug 2004

    Downloads:

    Abstract

    An efficient and transparent Grid resource management is a key element of any large scale Grid as the pure availability of a large amount of resources connected by a network does not constitute a Grid. As a Grid is dynamic by nature, it is essential to discover the available resources, coordinate the resources required by a complex job and monitor the progress of this job. All of this is subject to accounting and billing requirements, that is taking economic constraints into account. This complex and cumbersome process cannot be handled by the user directly. Therefore, a large-scale Grid requires an elaborate resource management system that hides this process from the user. However, the Grid resource owners have conflicting performance goals that have to be considered in a Grid management system. This tutorial presents an overview of the state of the art and experiences in Grid resource management and scheduling. In addition, an outlook is given on the design of future Grid management systems.

    Last modified: May 24, 2004