From: Mustafa Akgul (akgul@Bilkent.EDU.TR)
Date: Mon 29 Jul 2002 - 15:54:48 EEST
Date: Mon, 29 Jul 2002 03:52:33 -0700 (PDT)
From: Grid Today <grid@gridtoday.com>
Message-Id: <200207291052.DAA59314@gridtoday.com>
To: akgul@Bilkent.EDU.TR
Subject: 100140 BUILDING A COMMUNITY GRID - MY VIEW FROM THE TRENCHES 07.22.02
X-Virus-Scanned: by AMaViS snapshot-20010714
BUILDING A COMMUNITY GRID - MY VIEW FROM THE TRENCHES 07.22.02
By Fran Berman, Director, San Diego Supercomputing Center GRIDtoday
==============================================================================
For many of us who have been working in Grid Computing for the last decade,
the current environment is exhilarating. At present, there is strong interest
in the Grid from researchers, developers, funding agencies, the commercial
sector, and a growing segment of resource-limited applications communities. If
any time is the time for Grid Computing, this is it.
At the same time, there is a growing concern that the Grid has been oversold.
In particular, the concern is that the Grid will not be able to deliver to
applications developers and users its promise of immense resources, high
performance, and ubiquity. This concern is legitimate, as the Grid will
ultimately only be as successful as its user base.
The issue is not that the promise of the Grid has been oversold as much as
that the difficulty of developing the requisite Grid infrastructure has been
underestimated. Grid infrastructure is far from solely an integration project.
Serious and long-term research must be conducted to support the use of the
Grid as a virtual computing platform where an application can "plug in"
anywhere and achieve reasonable performance. An impressive list of problems
must be solved for the Grid to achieve its promise including scheduling of
Grid applications for performance as well as throughput, the development of
policies that can promote stability of the Grid and cross political, social,
and technological boundaries, the development of programming environments
which promote the development and deployment of adaptive,
performance-efficient programs for the Grid (such as those being developed on
the GrADS project), security, fault-tolerance, performance monitors,
debuggers, and other tools for the Grid, etc.
If we are thoughtful, serious, and organized about our approach to building
the Grid infrastructure, we should have a usable, useful and
performance-efficient Grid infrastructure by the end of the decade that
evolves with new hardware, new software and new applications. This is a lot of
work and much must be done in the mean time from infrastructure integration
and application development to basic research.
How will we build this vision of the Grid? Building the Grid and developing a
robust and usable software infrastructure requires coordination and
cooperation at an unprecedented scale. Given that Grid research started with
a relatively small community that has continued to remain closely knit (some
have said too much so), there has been an opportunity for agreement on a basic
service architecture approach and basic interfaces that have allowed the
community to work together effectively. In particular, a community model has
evolved for the Grid with a four-layer design.
The resource layer will always be whatever resources are available. On the
TeraGrid project, we have the opportunity to focus on the middle two layers
with a larger group because each site will deploy the same processor
architecture. With the Extensible Terascale Facility just proposed, the Grid
will be heterogeneous. Heterogeneity is a fact of life on the Grid and
provides both added capabilities and capacities to users while at the same
time making the process of developing the software infrastructure considerably
more complex. A large-scale, global, homogeneous grid is not an option. Facing
the heterogeneity problem early on will be required to help us gain the
critical experience needed to develop the Grid we have promised.
The next layer, common infrastructure, is especially important. No matter what
hardware underlies the Grid and no matter which applications are deployed on
the Grid, it is critical to have a layer of software that the community agrees
on.Global Grid Forum standards, the Open Grid Services Architecture, Globus,
NSFís Middleware Initiative (NMI), etc. are all part of the current picture
but the key part is agreement by the community about what services will be
available and how they will be interfaced. The Global Grid Forum, NSF
Middleware Initiative and other community efforts are key to the process of
achieving agreement.
The next layer is comprised of user-focused and community-focused middleware,
pre-NMI software, and other packages. While the common infrastructure layer is
evolving as largely open source, the middleware layer will likely have
proprietary software and vendor products as well as open source. In the
National Partnership for Advanced Computational Infrastructure (NPACI), we are
developing an interoperable collection of software called the "NPACkage" that
will be deployed at all NPACI resource sites on top of NMI and NPACI resources
(constituting an NPACI grid) as well as at other sites in the emerging global
Grid.
Finally, the applications layer provides the raison d'être for the Grid.
Without applications that truly find the Grid a useful platform, the Grid
cannot be successful. Note that if the Grid is developed properly, it will
provide access for applications to any of the Grid resources, whether or not
the application is distributed. It is critical for our community to work
closely at the outset with the applications communities to develop useful
application models for the Grid. The Grid user community of the next decade
must be a broad community encompassing existing Grid application paradigms
(embarrassingly parallel applications, parameter sweeps, staged applications,
etc.) as well as applications with dependences (such as those found in
computational biology), data-oriented applications (the "killer apps" for the
Grid over the next decade), applications with polyalgorithms (which can
adaptively target application algorithms to best utilize the available target
resources), etc.
The development of this Community Grid Model, as well as serious research into
the problems of developing the Grid that are not "low hanging fruit" is
critical to fulfill the promise of the Grid. The rapid development of todayís
Grids demonstrates that an enormous amount can be done in a relatively short
period of time. With serious focus, serious resources, and serious
cooperation, we should be able to achieve the full vision of the Grid and help
usher in a new era of science,software, and technology.
Fran Berman holds the title of Director for both the San Diego Supercomputing
Center and the National Partnership for Advanced Computational Infrastructure.
***************************************************************************
Full background information on all Sponsoring Companies
[ ] 921) SGI [ ] 934) Hewlett-Packard
[ ] 527) Intel [ ] 942) Sun Microsystems
[ ] 909) Fujitsu [ ] 539) Microsoft
For sponsorship information contact: gridads@gridtoday.com
GRIDtoday welcomes bylined comments for publication.
***************************************************************************
Copyright 2002 GRIDtoday Redistribution of this article is forbidden by
law without the express written consent of the publisher. For a
subscription to GRIDtoday, send e-mail to gridfree@gridtoday.com