DAILY NEWS AND INFORMATION FOR THE GLOBAL GRID COMMUNITY /    ( Table of Contents )

Special Features:

GETTING A DIRECTION IN THE SEA OF GRID
By Rich Wellner, Gridwise Technologies

As with many ideas in IT, Grid computing is one which was around for a while before being recognized by the mainstream. In fact, if the current crop of cluster tools and job managers sporting the name "Grid" is any indication, it will still be a while before the real benefits of Grid computing are realized.

When we speak of a "real Grid" we are referring to Foster's three point checklist. A Grid, as opposed to other distributed systems is one that:

• coordinates resources that are not subject to centralized control.
• using standard, open, general-purpose protocols and interfaces.
• to deliver nontrivial qualities of service.

Accordingly, many of the products currently available aren't quite yet Grid technologies. Most fail on the "standard and open" test, instead relying on proprietary protocols to build systems which lock users into a particular solution. Similarly, many fail on the "non-centralized" count. There are a few genuinely decentralized systems out that, but the bulk still have single points of failure.

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!

In a sense, this is a good thing. Often, the IT industry sets the bar pretty low and then declares success after leaping over a stack of match boxes. With Grid computing there is in its very definition an assumption of non-triviality. It does present a bit of a problem for managers who are interested in bring Grid technology into their shops.

First, let's review the reasons people move to the Grid:

• Make more efficient use of existing resources

• In many organizations there is a tremendous amount of IT infrastructure sitting unused at any given moment. This includes machines, networks and storage. If these resources can be captured and used effectively then an enterprise as a whole benefits.

• Lower barriers to productivity

• The Grid can provide productivity benefits along two axis. First is the ability of a user to more quickly get results from a given task. Second is the ability for a user to gain access to resources he otherwise would not be able to access conveniently in a conventional shop.

• Create an infrastructure for dynamic scaling

• The Grid, as in the monolithic Grid -- compare to the monolithic Internet -- will, and to some limited extent already has, create a new ability to call upon resources all over the world to execute a given task. In science this is being done today. The various physics Grids, for example, have resources all over the world and all these are mustered to the task of analyzing the way the world works. An international corporation can also take advantage of this kind of scaling by using resources in sites which are available during off-hours times for that site. For example, an office in Chicago requiring many CPU's to run a financial analysis might take advantage of the London offices spare cycles after their British counterparts have left for the day.

• Improve turn around time

• A Grid solution has the potential to utilize many, many more resources to solve a problem than a conventional system. One fairly obvious fallout from this is that a task which previously took a few days might be refactored such that many resources complete the same task in a few hours. This is perhaps the most obvious benefit of the Grid.

• Allow for new kinds of collaboration (both B2B and B2C)

• A persistent challenge for large enterprises is communicating with its partners and customers. The partner channel is particularly interesting because it has been historically been the domain of a handful of different companies with a handful of different standards. Not only were there no comprehensive security and protocol standards, but there was also a pretty enormous front end expense to getting collaboration systems in place. These costs were not only software and links (either dedicated line or VPN based), but also in requiring various staff members learn the intricacies of the new system.

Given the state of Grid technology, it is prudent for managers to evaluate which benefits they desire to exploit. In other words, they must compile their requirements. As with any other technology selection, all requirements are not created equal. A bank is typically more interested in latency and security than it is with aggregate bandwidth. A manufacturing facility operating a just in time assembly line will be interested in adherence to open standards. An investment firm might rank security and nontrivial services levels as their most desired Grid attributes.

Having prioritized those requirements it is then possible to embark on an evaluation of what solutions best match the combination of requirements and the three Grid criteria mentioned above. As examples, let us break down each of Foster's requirements and discuss how to go about evaluating tools. Bear in mind that the Grid market is fairly young and the point of this article is not to point at a particular tool, but to talk about the heuristics one can use to choose between a variety of less than perfect options.

Coordinates Resources That Are Not Subject To Centralized Control ...

There are a number of different tools on the market today that don't exhibit good decentralization of resources. The DNS system is a good example of a non-Grid decentralized system. Instead of having all domain name queries resolve to a host, there are peers of hosts around the world acting as the canonical sources for information. Someone searching for decentralized Grid tools can use this example when talking with vendors. Some tools provide no decentralization. Some provide fairly trivial mirroring for fail-over purposes. A handful provide fairly robust service decentralization.

Our view on the market is that the products ability to meet this criteria have more to do with the kinds of services being offered than the companies producing them. Some services are just flat out more easily distributed than others. The buyer should beware on this subject and ask pointed questions of their vendors regarding the specific services they will be deploying.

Using Standard, Open, General-Purpose Protocols And Interfaces ...

This subject is probably the hardest to attack in this era of the Grid's development. There are a number of open interfaces available from organizations like GGF, Oasis and W3C concerning how Grids should work. However, these standards haven't been very stable and thus many vendors have thrown up their hands and done their own thing for the sake of getting a product shipping. For intragrid work, these products are usually fine. In the case where a manager decides that he must have adherence to open standards for the sake of integrating his operation to the Grid -- as opposed to building an intragrid -- there are more limited choices.

Even when a Grid toolkit is selected which does adhere to open standards there can be integration problems of the most profound sort thanks to standards churn. In a young field like Grid computing the manager must make certain that a GridFTP server from vendor "A" is claiming adherence to the same version of standard that a GridFTP client from vendor "B" has implemented. Even larger failures will be possible in the next year as products migrate from OGSA to WSRF standards and are thus completely incapable of intercommunicating. Fortunately, the WSRF standards which are emerging have broad industry support and move the state of the art forward significantly. We anticipate acceptance to be fairly rapid, thus settling that part of the issue.

To Deliver Nontrivial Qualities Of Service ...

In isolation, this is the easiest of the list to satisfy. There are a lot of tools out there which can give managers very good bang for the buck if this is the only consideration. There are scalable databases, job managers and networking tools from a variety of different vendors which can bring nearly immediate results in terms of creating a system that is reliable, large and distributed.

What does all this mean for a manager trying to get a jump in this round of technology acquisition? At some level, it means the same thing it always meant, new technology is a risky business and a little education combined with a willingness to make a critical analysis of vendors is tremendously helpful. There is a difference between Grid technology and other idioms which have gone before it though. With the Grid, the possibility exists to do things in ways which haven't been explored before. Even in previous distributed computing paradigms, there have been only very low level standards available which provided pretty rudimentary capabilities. Sure, talented people could build great systems with them, but it was hard work. The possibility exists with the anticipated emergence of a global Grid analogous to the global Internet of providing levels of service impossible without all three legs of the stool.

About Rich Wellner

Rich Wellner has been a consultant based in Chicago for 15 years, working with a wide variety of clients in different industries. For the last decade he has concentrated on distributed systems, including over five years of experience in the Grid community. His Grid work has included high speed data transfer research, multi-Petabyte storage facilities and advanced job management systems. He is currently the president of Gridwise Technologies US, a Grid specialty consulting and development house with offices in the United States and Europe, where he works daily with organizations to advance effective use of Grid computing.

Wellner will be a panelist during the Grid Today 2004 discussion "Measuring ROI Across Grid Computing Environments" and will be available for conversations at the Gridwise Technologies booth in the exhibition hall.