Desktop Grids: Critical Systems and Applications Research (DGRID 2003)

Charles L. Brooks III: "Utilizing Large Distributed Computational Resources in Molecular Biophysics"

Abstract:

In this talk I will review our experiences in implementing large-scale computational problems in molecular biophysics on distributed resources. I will discuss the strategies explored for computations on the grid using Legion, Globus, the MMTSB Tool Set and related environments. I will conclude with comments on the potential for the development of successful distributable applications in a desktop grid environment.

Slides: acrobat
Link to the webpage of Charles L. Brooks III

Vijay Pande: Folding@Home: "New, Linearly Scalable, Large-scale Grid Algorithms to Break Fundamental Barriers in Computational Biology"

Abstract:

Modern computational biology has a nearly insatiable need for computational power, with major applications fundamentally limited by the lack of computational power. Traditional large-scale computational biology codes are tightly coupled, requiring supercomputers with large bandwidth and low latency networking. Desktop grids affords a great deal more computational power, but with the difficult barrier of adapting tightly coupled algorithms to the high latency, low bandwitdh capabilities of grids.
Here, I present our methods to decouple tightly coupled applications, allowing one to use loosely coupled desktop grids, with scalability to 10,000 processors. To test and facilitate these calculations, in 2000 we lauched Folding@Home, a distributed computing project in which over 500,000 CPUs have contributed. In particular, we have developed novel methods for the calculation of dynamics on long timescales (1000 to 1,000,000 times longer than previous calculations) as well as high precision free energy calculations, as one would use in the design of small molecule drugs.

Slides: acrobat
Link to the webpage of Vijay Pande

Philip Bourne: "High Noon - Bioinformatics versus the Grid"

Abstract:

Why High Noon? There are millions of VCR's in the US alone that are flashing "12:00" right now. It is not that their owners cannot set the time or program them, it is just that the barrier to entry is too high. It is more trouble that it is worth. I will argue that today the same is true for the large scale application of the grid to problems in bioinformatics - it is more trouble than it is worth. We and others are addressing this situation by first understanding how science is really done and what grid technologies need to be mapped to that process. Pragmatically that leads us to developments in work flow, understanding the data such that, for example, redundancy can be exploited and new approaches to the visualization of results. Examples from our research and that of others in these areas will be discussed.

Slides: acrobat
Link to the webpage of Philip Bourne

Franck Cappello: "Ontology of Desktop Grids: a Pragmatic View from the XtremWeb Experience"

Abstract:

Despite the experience acquired with SETI@home, Distributed.net and the other large scale projects, we still have a limited understanding on key issues related to multi-users/multi-applications desktop Grids.
Taxonomy of Dgrids is not so easy to establish. The classification could certainly consider the system architecture and its main application. It should also consider the deployment/usage parameters. We also face the difficulty that the application domain of DGRID is not well defined yet.
Who can you trust if you are a CPU cycle provider, a client or if your role is to control/coordinate the system? Depending on the authentication, tracing and coercion/certification capacities of the system environment, techniques like sandboxing, end to end parameter/code/result encryption and result certification need to be used.
The programming model properties are certainly a strong acceptation criterion of DGRIDs. Should we use classical and widely used programming libraries like MPI and RPC, even though they do not provide any means for handling the intrinsic volatility of DGRIDs nodes, or should we consider/design more suited but also more exotic/restricted programming approaches?
In this talk, we present our view about these issues acquired during the XtremWeb and MPICH-V projects. We especially detail the third issue, related to programming models.

Slides: acrobat
Link to the webpage of Franck Cappello

Miron Livny: "If you can do it on the Desktop you can do it everywhere"

Abstract:

Slides: acrobat
Link to the webpafe of Miron Livny

Adam Beberg: "Distributed Computing without a Supercomputer"

Abstract:

GRID computing standards have evolved down from the world of researchers with supercomputers, so it's no big surprise that these systems have not quite fit with the desktop and corporate world. And those models do not apply to the end user world in anyway way. Distributed capabilities in user applications need to not only be seamlessly integrated into the applications, but have to be useable without administrators or 3rd party intervention.
With this in mind, we will take a look at the applications that can benefit from this technology, and what has to happen in the developer world to make it simple to add distributed functionality into new and existing applications.

Slides: acrobat
Link to the webpage of Adam Beberg

Last Change: Aug 2003
Author: Taufer Michela