 |
|
DAILY NEWS AND INFORMATION FOR THE GLOBAL GRID COMMUNITY /
|
Special Features:
GRID: A WORK IN PROGRESS WITH $4B POTENTIAL
By Derrick Harris, Editor
This week (May 24-26) at Gt'04 in Philadelphia, leaders in the Grid community
-- including HP, Sun, IBM and Intel -- will gather to give valuable
information about the use of Grid technology in the enterprise. To learn more,
visit www.gt04.com .
The promise of the adaptable, always-available Grid computing looms large ($4
billion market projections by 2008 abound) but how do today's managers execute
a distributed computing strategy designed for current business problems? What
lessons about Grid deployment has HP learned from its work with leading edge
customers and partners? What new technologies exist to help secure and manage
distributed computing resources? GRIDtoday recently sat down with HP's
marketing manager for Grid computing, Sara Murphy, to find out how more about
managing Grid computing expectations.
GRIDtoday: What are the business issues driving Grid computing?
SARA MURPHY: The primary business drivers fall into three main categories: 1)
Economic pressure to do more with less; 2) Geographic and organizational
distribution of teams and computing resources; and 3) The need to adapt
quickly to change. Let me elaborate:
- The amount of IT gear is significant. Servers, storage, PCs, laptops,
printers, switches, routers, PDAs, are filling up data centers and offices.
With this volume, comes complexity and difficult management and utilization
issues. One of the most immediately obvious benefits of Grid technologies is
the ability to take advantage of underutilized resources. Most organizations
have peaks and troughs in demand. For example, while it is sleep-time at one
organization's data center, it is prime time at another. Spreading the
computing load across organizations and geographies provides better
utilization, which improves ROI.
- Centralization is unsustainable. Co-located resources managed by a single
central authority will eventually, if not already, become unsustainable.
Multi-national companies are not centralized. Corporate data centers are, of
necessity, not all in one location.
- The ownership of the content of science, commerce and consumers is
dispersed, but not necessarily public. It resides in geographically removed
locations, within separate ownership domains and separate management domains.
However, if it is not accessible to collaborators, partners, customers and
colleagues then it has little value.
- Geographic and organizational dispersion of virtual teams is increasingly
common. We need to take advantage of it. Very few of us only interact with
colleagues in the office next door. We collaborate with colleagues from other
parts of our own companies as well as with partners, suppliers and customers.
Gt: Do you think Grid has been over-sold? In a recent GRIDtoday interview with
Alan Blatecky, the new RENCI deputy director, he commented that the "hype
surrounding Grid computing is becoming destructive, as Grids have tended to be
oversold both in terms of when they will be available and what they can
actually do. As a result, one of the largest obstacles facing realization of
the Grid vision is the inability to manage expectations." Do you agree?
SM: Yes, I do agree. Grids do not "snap" together and aren't operating in
every corporation. We are seeing customers who believe they "must have" a Grid
when a cluster would solve their problem. That's why the assessment of the
computing environment is so important and why HP has dedicated consultants
worldwide to help customers get Grids up and running. Not all distributed
computing solutions require Grid technology. Let me give you some examples:
1). If you want to share and use homogeneous computer equipment in a single
data center, you probably need a cluster and not a Grid. Cluster solutions are
scalable systems built to leverage speed, flexibility and throughput -- but
there are physical boundaries (interconnects, geographical) associated with
clusters, even when the cluster infrastructure is capable of supporting
thousands of processors.
What can get confusing is that a high-performance supercomputer or cluster can
serve as a node on a Grid. This is the case at the U.S. Department of Energy's
Pacific Northwest National Laboratory (PNNL), one of the world's premier
molecular and life sciences labs. The PNNL 9.2 teraflop HP supercomputer is
connected to the DOE Science Grid and the supercomputer is the largest node on
that Grid.
2). If you want to share and use heterogeneous systems in a single data
center, the connection solution might be the Utility Data Center, which also
could serve as a node on a Grid. The Utility Data Center (UDC) is a
self-adapting, self-healing and policy-driven system in which computing assets
are wired once, then provisioned virtually and automatically. The UDC lets
organizations allocate and reallocate resources on the fly.
3). A business situation calling for productive use of all IT resources --
computers, printers, storage space, software -- within a common firewall yet
across multiple geographic locations, will require Grid technology not
clusters. The goal of Grid computing is to treat all IT resources as services
that are available on demand in a heterogeneous environment. Plus, Grids can
and do cross organizational boundaries.
4). Finally, if your business goal is to create an environment in which you
can share resources globally and collaborate across corporate firewalls with
partners and suppliers, you are looking for true Grid. One example of a
leading edge Grid application is our collaboration with BAE Systems Advanced
Technology Center and several leading educational and research institutions.
These partners are exploring how to use Grid in advanced, collaborative
simulation and visualization in aerospace and defense design.
One misconception about Grid is that it is restricted to a particular class of
applications, those that are "embarrassingly parallel." Grid computing is
about running application workloads on a pool of shared resources. Scientific
workloads that are easily parallelized are a particularly good fit to this
model but not the only appropriate use for Grids.
Grid computing also is driven by new and demanding applications, such as the
three-year Adaptive Enterprise @ Singapore collaboration HP has with the
Infocomm Development Authority of Singapore (IDA). This collaboration is using
the deployment of Grid and utility computing technologies to spur new growth
areas such as online games and digital media.
Gt: What are the key issues an HPC manager must evaluate when considering Grid
computing?
A: Every Grid assessment is different. I guess that's lesson number one. Other
key areas to focus on:
1). Recognize the complexity of the task. An enterprise Grid must be smart
enough to navigate boundaries, such as corporate firewalls and networks, and
reliable enough to process mission-critical applications. That's quite a range
of technical prowess, especially when it comes to delivering Grid to the
enterprise space where the reality of Grid is three to five years out. In
high-performance computing, we are just starting to see the power of Grid. For
example, HP has joined forces with the University at Buffalo, State University
of New York (UB), to deploy infrastructure technology and academic resources
that will power the university's world-class Center for Computational
Research. As a foundation for bioinformatics and life sciences research, HP
and UB are building an open storage area network (SAN) with a capacity of 75
terabytes -- capable of housing roughly four times the information found in
the Library of Congress.
2). Acknowledge the business advantages. The economic promise associated with
Grid is outstanding. The IT function, whether in an engineering or commercial
enterprise, is accountable for driving down costs, creating new value, making
IT as a service more agile, secure, reliable, flexible and adaptable to any
sort of change. In today's business and research environment, the only
constant is change and we believe Grid technologies will be key contributors
to enterprise adaptability.
3). Invest in and support open standards. Open standards drive Grid
deployment. Imagine what would have happened to the Internet if it had been
built on proprietary, non-interoperable technology? The only way that Grid
will be able to provide the universality it offers is if it is built on open
standards.
4). True Grid is a community, not one company, play. The Grid requires
technical innovation from a community of companies and thousands of
developers. The collaborative research underway mirrors the very promise of
Grid computing. For example, research is underway to develop data-intensive
Grid technologies that will be used by the worldwide community of scientists
working at the next-generation Large Hadron Collider (LHC). HP is supporting
an operational Grid for the LHC at CERN, the European Laboratory for Particle
Physics. The LHC, the world's largest scientific instrument, enables research
into the fundamental nature of matter. It is in the final stages of
construction at CERN's facility outside Geneva. HP will link computing
resources at its HP Labs locations in Palo Alto and Bristol (United Kingdom),
as well as HP Brazil and HP Puerto Rico to CERN's LHC Computing Grid (LCG) to
help manage and analyze the massive quantities of data expected to be produced
by the facility. The opportunity to participate in the LCG will provide us
with unique insight into the functionality and complexity of large-scale Grid
environments.
Numerous software companies are working with us to help make Grid ready for
prime time. For example, HP partners with Altair Engineering, Avaki, Oracle,
Platform Computing, United Devices and others to develop the full potential of
Grid technologies. Ultimately, Grid customers around the globe will reap the
benefits from these collaborations.
5). Understand and address the cultural and people issues: There is a tendency
for people to hug their own resources and not want to share. There has to be
an advocate in the organization to encourage sharing because it is often
difficult to give up control. The issue of management and accounting comes
into play, too. As people share resources, they want to account for them
effectively. This is especially true if a Grid solution spans multiple
departments because each department needs assurances that it is getting as
much from the Grid as it is contributing. Ultimately, the Grid should render
computers, processing power, data, Web services, storage space, software
applications or devices as Grid services.
Gt: Tell me about some of the latest innovations in Grid technology coming
from HP Labs.
A: HP's primary focus is on the management and execution of Grid services.
We've made an ongoing commitment to "Grid-enable" our entire product line,
from the smallest hand-held devices, PCs and printers, to the largest servers.
We have ported the Globus Toolkit, the open source Grid infrastructure
implementation, to all of our server platforms.
Grids are necessarily complex and dynamic environments and the ability to
manage them and effectively deploy applications is vital. To tackle these
challenges, we are building on the OpenView platform (the infrastructure
management platform used by every single Fortune 50 company but one) and
delivering technologies to address some of the show-stopper issues related to
Grid deployment. These technologies include:
1). Web Service Management Framework (WSMF): HP's WSMF was designed to manage
Web services using Web services. It is being refactored to do likewise for
Grid services. WSMF allows for management, monitoring, control, metering and
event handling for infrastructure, applications, Grid services and business
practices. WSMF has been submitted to the Web Services Distributed Management
(WSDM) working group of the Organization for the Advancement of Structured
Information Standards (OASIS) body as the basis for the WSDM standard, which
is still under development.
2). SmartFrog: HP Labs invented a software technology for describing
distributed software systems as "collections" of cooperating components, and
then activating and managing them. The core SmartFrog framework is released to
open source (LGPL or Lesser GNU Public License). Also, SmartFrog is the basis
for work being done by the Global Grid Forum's (GGF) Configuration
Description, Deployment and Lifecycle Management (CDDLM) group.
3). Automatic Flexing Interface: HP recently announced a new technology,
called the Automatic Flexing Interface (AFI), that allows customers'
applications to control the "request" and "release" of compute resources in a
data center. This application programming interface, developed by HP Labs, is
based on Grid standards. Utilizing industry-standard software protocols, AFI
simplifies integration and management of a heterogeneous infrastructure,
including Grid deployments.
4). Topology designer: HP Labs has developed a software interface to enable
the design of Grids via an open, easy to use GUI. This reduces the complexity
of defining and managing Grids.
5). Utility Data Center or UDC: The Utility Data Center is a wire-once,
dynamic, programmatically configurable, data center. The UDC allows
organizations to allocate and reallocate resources on the fly, transforming
the data center from a static repository of applications and data into a
dynamic computer-power generating facility. The marriage of UDC with Grid is
particularly powerful. If a UDC is presented with a request for a Grid service
that it cannot currently meet, it can programmatically reconfigure itself to
meet the request. For UDC and Grid, the whole is greater than the sum of the
parts.
Gt: Sara, you have been involved in distributed computing from the Digital
Equipment days. All this Grid and cluster discussion must seem old hat to you.
What's different?
SM: Our high-performance heritage does include years of expertise in the
distributed computing area. HP and Digital both got their feet wet on utility
computing/time-sharing back in the 1960s. And in the 1980s, Joel Birnbaum, at
the time HP's Vice President of R&D, advocated the notion of utility computing
where IT services would have attributes similar to those we expect from
utilities providing electricity, light or clean water. Now, almost 40 years
later, we are working with customers who want to build Grids that connect
thousands people scattered across the planet.
What's exciting about Grid computing is that it has the potential to enable
resource sharing and collaboration across the enterprise and across multiple
organizations. Grids can solve real computing problems by simplifying global
access to enterprise computing services. Grid computing has the potential to
accelerate innovation and creation. Corporate customers, engineers and
scientists need to deploy resources quickly to respond to new market demands.
It's easier to be agile when you can tap into resources over a Grid -- not to
mention more cost effective.
The drive toward this model is unstoppable. For example, HP Labs just finished
working with DreamWorks on the soon-to-be-released film, "Shrek 2," and
provided the scalable, off-site rendering capacity for the production. The
Labs' data center became a remote extension of DreamWorks' IT infrastructure,
providing the computing boost needed for peak periods in the production
process. The data center that resulted consisted of 500 HP servers (1,000
processors) connected to DreamWorks' studio 20 miles away in Redwood City,
Calif., via a secure fiber optic link. It is an example of HP's Adaptive
Enterprise model for computing as a service, in which companies can draw on
expanded computing resources when they need it most, without having to
purchase or manage physical computing assets.
Over time software, storage, server and IT service businesses will change to
adopt this model -- driven by the demands of their customers. Of course, I
like to remind people that Grid computing is still a work in progress.
Web services, utility computing, .NET, CPU harvesting and distributed
computing are just a few of the technologies that fall under the Grid
computing umbrella. Gt04 -- a premiere enterprise Grid computing conference
targeting industrial and commercial users -- will gather experts, and outline
strategies and road maps for Grid deployment. For more information, visit
www.gt04.com .
Grid computing is here!
|