That means all we really are talking about is scale. Your enterprise machines already are connected via TCP/IP; they probably have excess storage to share; and we know you have a few CPU cycles to share. Lets get realwe have lots of resources to share. I have seen estimates that most office PCs are about 95 percent underutilized. Now I know the real cost of hardware ownership is decreasing constantly, but there is a wealth of potential savings and/or revenues to be gained by really utilizing the computational power we have paid for. Whats the catch? The catch is this: Is the cost of implementing and maintaining a grid computing system greater than the potential savings? At the end of the day it really may be more efficient to have multiple stand-alone machines that perform specific tasks than to enable a distributed virtual machine.

What Its Not
Large peer-to-peer systems, such as Gnutella, which enable access to files on servers whose only commonality is they have IP addresses, often are considered a limited form of grid computinglimited in that all the system is sharing is files. Real computing resources are not shared. There also is no central controlling authority. A true computing grid is more than peer to peer, because more than just files are shared. Likewise, the Internet is not a computer grid. Real grid computing provides on-demand computing systems using controlled connected resources.

A Perfect System
Exactly what would constitute an ideal grid system for the enterprise? It must:
Make use of existing infrastructure.
Have no discernable effect on a users machine (that is the machine whose resources are being reallocated to the grid).
Demonstrably improve productivity and time-to results.
Have adequate security.
Provide instantaneous access to computer and data resources.
Be seamlessthe user (of the grid application) need not even be aware where and what computing resources are being consumed.

Think of grid computing like a power grid. Electricity isnt the big dealdistributing it is. Likewise a computer isnt a big dealefficient distributed computing is. (Wow, maybe Larry Ellison really was rightthe network is the computer.)

Underdeveloped countries have generating plants and electricity. What they dont have is an efficient, fail-over, redundant, seamless power grid like we do in North America. After Hurricane Mitch in 1998, some major regions of Nicaragua were without power for six months. In the U.S., electric power is always (well, almost always) available because of sophisticated systems that distribute power to where it is needed, when it is needed. Power is shared seamlessly over the entire grid in such a way we really dont know or care where it was generated. Just in time, peak-demand power stations start up and dump power onto the grid when necessary. How is that accomplished? It is accomplished with a solid infrastructure and a complex and reliable system that manages distribution of power over that infrastructure.

We already have an existing solid infrastructure in our enterprise computing systems. We have modern, efficient PCs, workstations, and servers connected over redundant TCP/IP networks with backup power sources and data backup or redundancy. All we need to create a computer grid using that infrastructure is a control system (i.e., software).

Standards
Like most everything we do in this business, we must have standards for implementation and development. Such a standard existsOpen Grid Services Architecture (OGSA). The OGSA is a work in progress presided over by the Global Grid Forum (GGF, www.ggf.org/). The GGF is a community-initiated forum of 5,000-plus individual researchers and practitioners working on distributed computing, or grid technologies.

In this country the Globus Project (www.globus.org) is the most visible and active organization. Globus was started in 1996 at Argonne National Laboratory in cooperation with the University of Southern California and the University of Chicago. The Globus Project is a research and development project focused on enabling the application of grid concepts to scientific and engineering computing. Its deliverable is the Globus Toolkit, an open-source grab bag of tools designed to enable implementation of grid computing.

The Globus Toolkit
The Globus Resource Allocation Manager (GRAM) provides resource allocation and process creation, monitoring, and management services. GRAM implementations map requests expressed in a Resource Specification Language (RSL) into commands to local schedulers and computers.
The Grid Security Infrastructure (GSI) provides a single-sign-on, run-anywhere authentication service, with support for local control over access rights and mapping from global to local user identities. Smartcard support increases credential security.
The Monitoring and Discovery Service (MDS) is an extensible grid information service that combines data discovery mechanisms with the Lightweight Directory Access Protocol (LDAP). MDS offers a uniform framework for providing and accessing system configuration and status information such as compute server configuration, network status, or the locations of replicated datasets.
Global Access to Secondary Storage (GASS) implements a variety of automatic and programmer-managed data movement and data access strategies, enabling programs running at remote locations to read and write local data.
Nexus and globus_io provide communication services for heterogeneous environments, supporting multimethod communication, multithreading, and single-sided operations.
The Heartbeat Monitor (HBM) allows system administrators or ordinary users to detect failure of system components.
The toolkits provide APIs for all services written in C as well as command-line tools. Most major services also are available as Java classes for implementation on any operating system with a Java VM. One nifty feature of the Globus services is they can be implemented incrementally. You dont need to create a full-blown enterprisewide grid to get started. You can start small and grow the grid (and the grids functionality) as needed.

Its Not Just the Technology
The scientific community will have no problem embracing grid computing. Intel and Oracle already have massive grids in place for microprocessor development and software testing, respectively. IBM (which is a major grid proponent) eats its own dog food with its IBM intraGrid based on the Globus Toolkit. You expect physicists counting leptons or neutrinos will rely on grid-enabled supercomputers. But you may have a difficult time creating an effective grid in the modern corporate enterprise. Why? Because we first must deal with people, and people will provide the real stumbling blocks.

Resistance to change and territoriality are major problems in all large organizations and become more difficult to deal with in organizations that are divided into silos of middle-management mafias ruled over by highly compensated Dons (I mean senior managers). Sharing re-sources between spheres of power and/or influence is a touchy subject. There will be a lot of the NIMBY (not in my backyard) syndrome: While I agree making the best use of our available computer resources is an excellent concept, my servers are off limits. . . . For grid computing to be effective, it must be driven from the top down.

Into the Future
This stuff is all brand new and still being developed. The Globus Project isnt even a decade old. I suspect the scientific and academic communities will do a lot of the dirty work for the rest of us. It wont be long before implementing computing grids will be relatively simple and inexpensive. Of course, that doesnt mean it will be the proper business decision. Infrastructure may become so commoditized that attempting to use all available resources wont make bottom-line sense. In a larger aspect, though, it does make sense. We live in a world dependent on ever-dwindling supplies of natural re-sources and an insatiable demand for power. There is a certain beauty in making the best use of all of our resources, even if they are only CPU cycles.