Microsoft gets on the grid

Microsoft is creating a "Cluster Compute" version of Windows and intends to work more closely with grid industry standards bodies, Tony Hey, the company's corporate vice president of technical computing, said in an interview with CNET News.com on Tuesday.

Grid computing is a vague term that describes ways to make several machines work together to efficiently tackle computing jobs. The reason to use grids is to get more bang for the buck--by letting multiple applications share formerly separate resources or by automatically juggling priorities.

Microsoft's competitors, such as IBM, Sun Microsystems and , have each spelled out their plans for grid computing which, in general, rely heavily on the use of Linux on low-cost servers. Grid computing is still largely the purview of scientists and researchers, but to make better use of their hardware.

Yet despite the growing interest in grids, Microsoft has been largely quiet on the topic apart from internal research efforts. And perhaps not surprisingly, its vision differs somewhat from that of existing grid vendors.

Hey, who joined Microsoft in June after heading up the U.K.'s e-Science research initiative, said that Microsoft intends to focus its development efforts on so-called data grids, rather than compute grids, as many companies and standards bodies are doing.

Data, or information-centric, grids are designed to let people access information from disparate sources and then combine that data as part of a computing job.

For example, a hospital could view X-ray information from its own imaging system and simultaneously access relevant patient data from different locations. Or a financial services company could do data mining analysis on multiple databases without having to send large amounts of information across the network.

"Federating data--that's the big payoff. Aggregating compute cycles to me is the easy piece," Hey said. "You won't want to move petabytes of data around. You can move (server) clusters to the data."

Microsoft is focusing on sharing information--rather than computing power--through grids because that reflects what the academic and research communities are wrangling with, Hey said. For example, researchers could cull data from sensors with satellite data when predicting weather.

He said many of the grid technologies used in academia can be applied to business applications. Hey spoke at the GridWorld conference in Boston on Tuesday.

The hiring of Hey and Microsoft's investments in high-performance computing versions of Windows are signs that the company is taking the grid software market seriously, said Songnian Zhou, chief executive and founder of grid software company Platform Computing.

"The reality is they don't want Linux to take over the world in high-performance computing," Zhou said. "Tony (Hey) is serious about being a good citizen in the grid community, where they haven't been an active participant."

But Illuminata analyst Jonathan Eunice was skeptical that Microsoft would embrace open standards to the degree that other vendors do.

"Tony's endorsement of open standards is quite interesting and I think significant, but I don't think it's a statement you can trust. It's open as far as open benefits Microsoft," Eunice said.

Eunice said a good first step to showing its commitment to standards would be creating a good implementation of the Globus Toolkit (standards-based software for building applications that can run a grid of several machines) on Windows, something the company has not committed to doing.

Incredible outlier
Microsoft's main product in the grid computing area is Windows Server 2003 Compute Cluster Edition, which aims to be a more viable alternative to Linux. At the company's Professional Developers Conference in September, Microsoft released a beta version of Windows Compute Cluster Solution, which is tuned for 64-bit processors and includes a job scheduler.

In another indication of the company's commitment to high performance, Microsoft Chairman Bill Gates in November will deliver the keynote speech at the SC/05 supercomputing conference in Seattle, which is

At that time, Platform Computing and Microsoft will announce a partnership, said Platform's Zhou. Platform Computing, which does not compete directly against Microsoft, provides grid middleware to schedule the movement of computing jobs among computers and set policies for how to divvy up jobs.

Although he declined to offer many specifics, Zhou said that the partnership with Microsoft will allow customers to use Windows desktop machines in a computing grid and have Windows clusters to interoperate with Linux servers.

Indeed, interoperating with existing Linux and Unix systems is required for Microsoft to make a bigger impact in the world of high-performance computing, said Microsoft's Hey. As such, Microsoft intends to be a more active participant rather than a "bystander" in standards bodies, such as the Global Grid Forum, where it has been a member for some time.

"Microsoft is now wanting to engage with the scientific community in open, interoperable standards. We want to see if we can do technical computing on Windows," he said, adding that he expected commercial grid software from Microsoft to coexist with open-source products.

Greater maturity in standards development and usage will help drive the use of grids, according to some industry executives and analysts. There are a plethora of grid-related standards, some of which are not fully developed or widely adopted, they said.

"We see grids where Linux was five to seven years ago or where the Internet was 10 years ago," said Ken King, IBM's vice president of grid computing. "We're trying to get to an open approach, and we'll apply pressure points (to grid software vendors) to get them to support standards."

IBM on Monday announced a partnership with grid software company Univa to include Univa's Globus Toolkit software on IBM servers.

The open-source organization Globus Alliance creates implementations of the Globus Toolkit, much the way the Apache Foundation creates open-source products, like its Web server, based on industry standards.

Instead of Globus, Microsoft's Hey is focusing on a data-access standards called OGSA-DAI, for Open Grid Services Architecture Data Access and Integration. The goal of the effort is to build middleware that can access disparate data sources from a grid.

Microsoft's stated focus on data grids, which are more difficult to build than compute grids, is another indicator of its motives, Illuminata's Eunice said. He added that Microsoft is an "incredible outlier" in the high-performance market where there is 25 years experience with Unix and Linux systems.

"It's a way of punting into the future and denigrating what's deeply, widely deployed in grid industry already," he said. "Essentially it's a delaying tactic to delay considering the Globus Toolkit or what have you and punt to something where there really aren't standards."

Johnson & Johnson's Pharmaceutical Research & Development division used proprietary grid software from United Devices to build its computing grid two years ago. But standards support is increasingly important, said Jeffrey Mathers, director of strategy and delivery at the division's technology office.

"The most important standards are the ones that get (packaged application) software to work on grids," Mathers said. "We buy from ISVs (independent software vendors) so it's important how they grid-integrate their software."