Globus Toolkit 4 Broadens Choices, Challenges Teams
"GT4 implements the latest standards from WS, OASIS and W3C," explained Globus Consortium board member Ian Foster, associate director of math and science at the Argonne National Laboratory, during our conversation late last week: "Its secure, and it has much higher code quality and testing after a six-month beta program with people on both the research side and the commercial side."
Also in that conversation was Mark Linesch, chairman of the Global Grid Forum, who emphasized the role of GT4 as a foundation for further work. "Developers can align with it, users can adopt it: It occupies a unique space in the grid ecosystem because its open-source nature invites adoption and imitation," Linesch said. The Globus Toolkit Public License (GTPL) is based on Version 2.0 of the Apache license.
Sun Microsystems, a member of the Globus Consortium, has put forward considerable amounts of its own code for open-source grid development including the Sun Grid Engine. I asked Sun grid-computing spokesperson Bjorn Andersson, another participant on our conference call, what criteria an enterprise could use to decide where a grid was a better approach than a large shared-memory multiprocessor system such as the massive Sun Fire E25K machine. "It comes down to looking at what type of application you need to run," he said. "If you have what we might call a stateful application, such as a large database, that will run on one of the larger SMP systems; there are other applications that will be more embarrassingly parallel that will run well on a grid."
The key point, Andersson said, is that broad adoption of vendor-neutral standards such as Web services reduces the need to force-fit an entire enterprise application portfolio into a single hardware architecture. "We have several places," he said, "where customers have high-end SMP systems and complement those with racks of smaller systems, and run that all on a grid." That same point was reinforced by Sarah Murphy, grid marketing manager at Globus Consortium member Hewlett-Packard, who said in the same conversation that "grids make resources more accessible, easier to discover and utilize: You can have a very broad offering and users can pick the system that best solves their problem. A supercomputer might be best to solve a very tightly coupled problem; a grid might better solve another part of the problem."
I asked the group how enterprise architects should distinguish a likely project from the kind of thing that their "skunk works" teams have likely been already doing with inexpensive building-block technologies such as Beowulf clusters. Whenever I write about grid computing, someone almost always writes back to me that many projects flying that flag are really just clusters: that the essence of the grid is the ability of resources to come and go, and for workloads to discover and use resources rather than requiring tasks to be designed and deployed on known and often fairly homogeneous cluster configurations.
"My interpretation is that in the Beowulf days, it was a task for a handful of people," responded Wolfgang Gentzsch, managing director of Grid Computing & Networking Services at the Microelectronics Center of North Carolina. "To build a grid, you have to build your community first and discuss how you want to move ahead. You have to do this very carefully and evolutionarily."
The Globus effort is certainly aimed at an international community, not only in terms of vendor interoperability but also in terms of international application support and other key technical details: "Our second-largest download volume is to China; half of the top five are Asia and Pacific," noted Globus Consortium President Greg Nawrocki.
Affordable, interoperable and increasingly developer-friendly grid technology intensifies both the opportunities and the competitive challenges facing enterprise development teams.
Tell me what it takes to bring you into the grid community at firstname.lastname@example.org.