OS/400 Platform to Participate in Grid-Based Web Services

by Timothy Prickett Morgan

Grid computing isn't a single technology but a spectrum of technologies. Everybody knows about the SETI@home screen saver, which is used by scientists involved in the Search for Extraterrestrial Intelligence to aggregate the unused processing capacity on desktop, workstation, and server computers to sift through radio signals coming into the Arecibo Observatory, in Puerto Rico, to look for alien life. (I have a dual-processor Dell workstation connected to a DSL line, and I would be happy to donate my excess processing capacity to the cause. Maybe my machine will be the one to find a radio broadcast from another planet.…)

Sun Microsystems got into grid computing when it bought a little software company called Gridware, in July 2000. Gridware provided software only a little more sophisticated than the SETI code but had the same basic idea. Sun's Grid Engine software, which runs on Solaris and Linux, taps the unused capacity on networks of workstations and servers and uses it to run parallelized supercomputer applications that might otherwise have to be run on a real--and real expensive-- supercomputer or cluster. Having sold lots of excess server capacity to companies in the late 1990s, Sun knows all too well that companies have all this processing power sitting around on machines that are running at 15 to 20 percent of capacity at most. Grid Engine aggregates that capacity and lets people use it. Sun has even recently merged Grid Engine with its iPlanet Portal Server, which allows companies to give users access to this capacity for running number-crunching applications, via a Web interface, from anywhere in the world.

Grids will be based on a bunch of different software and hardware technologies, but one of the emerging bodies controlling the evolution of grid computing is the Globus open-source project, which IBM has gotten behind in a big way in the past year. Microsoft also recently threw a $1 million bone to Globus to make sure there will be a Windows version of the open-source Globus toolkit software available. SETI might be famous, but the real opportunity is for grids based in companies, not in universities and government research institutions. These grids will be based on clustered servers and workstations, and they will present a single, global view of all the resources available on those grids. The machines in these grids will be configured in such a way that they can do stand-alone, monolithic work (like OS/400 servers do today) and grid-style number-crunching work (like they might assist with tomorrow). My sources at IBM say that grid computing might generate as much as $1 billion in revenues this year, and that it could grow to be as much as $4 billion to $5 billion in the 2004 to 2005 timeframe. Any market that is growing by $1 billion a year is going to attract attention, and grid computing is getting a lot of attention.

There will be public grids and private grids, local grids and ones that span the planet. The important thing to remember about grid computing is that the grid architects are trying to create a single environment in which multiple applications can reside and share processor, memory, storage, and bandwidth resources, to run technical applications and deliver a Web-based interface to users. This is a lot more difficult than creating a monolithic server to run a few applications; letting applications roam around a grid with their data to get the processing and bandwidth they need without stomping on other applications is not trivial; it is, in fact, a dream of science fiction. It is also a dream of business people who, as users, want cheaper processing capacity than they can get under the monolithic model or who want to set up grid utilities to supply processing capacity as a business. There is an elasticity in processing, memory, storage, and I/O capacity: The cheaper it gets, the more people use, and the more dependent they become on that capacity.

Fortunately, the key players in the IT market have realized that they have to solve many of the same problems for grid computing as they do in order to allow customers to create applications that follow the Web services model. This is an oversimplification, but Web services means creating distributed applications based on open protocols that reside on Internet-like software stacks that allow users to access these applications from anywhere on the Internet or from within intranets and extranets. Just like a grid application makes a collection of time slices on servers and workstations look like a supercomputer, a Web services application will make a collection of dozens or even hundreds of servers, perhaps located at many different physical companies, feel and behave like the monolithic data center applications that run on OS/400 and other midrange platforms today.

A few weeks ago, IBM and researchers associated with the Globus project announced a specification for a new set of grid standards called the Open Grid Services Architecture. OGSA seeks to merge grid and Web services standards so these two worlds of computing don't invent incompatible but functionally equivalent models. This spec is just getting rolling, and Microsoft is putting some of its weight behind it as well. IBM says it is committed not only to making the Globus toolkit available for its xSeries servers running Linux and its pSeries servers running AIX but also to eventually porting the Globus toolkit to its zSeries mainframes and iSeries midrange servers. IBM wants the Globus software to run across all four server families in its eServer product line because it wants all of its customers to be able to plug their machines into grids. Of course, knowing IBM, the Globus support will be enabled through Linux or AIX partitions on the iSeries. Weaving grid applications into OS/400 and DB2/400 is the hard part.

This may sound unnecessary or even like a luxury, but consider this simple scenario. A midrange manufacturer works with the partners in its supply chain to create a grid, to help itself and its partners design their own parts and the final product. Separately, neither the manufacturer nor its partners may be able to afford such a system. Also, because the core applications are running on servers that are grid- enabled, their manufacturing systems and databases can be plugged directly into the grid design systems. A grid-based system might also be used to look for sales trends in a data warehouse or to create an economic model for a particular market. In either case, security and authentication are important issues, just as they are for big government institutions using grids to do weapons and physics research today.

Before this grid-based Web services dream becomes a reality, we have to get through the hard work of creating and adopting a standard like OGSA. And it has to be implemented in code and used by real companies doing real work. Being an open standard that is a superset of the Globus toolkit, which is based on open-source code, will probably help the OGSA standard move along. In the meantime, my IBM sources say the company is working to make the Globus toolkit compliant with the J2EE standard, which means it is not yet suitable for supporting even rudimentary Web services applications. IBM also expects to have grid-enabled versions of its WebSphere application server and other WebSphere middleware in beta by then end of 2002. Production versions of grid-enabled WebSphere might be available early in 2003, or they may take a little longer.