Bristol-Myers Squibb Taps Grid Computing

You can't see it from the country road in western New Jersey, but in a brick building behind a dairy farm and across a corn field, a group of Bristol-Myers Squibb scientists and technology project managers are working on one of the largest "grid computing" systems in the world. And the system just might help the pharmaceutical giant boost profitability.

Tapping the company's internal high-speed network and deploying a package from software vendor Platform Computing, the Bristol-Myers team went live in mid-August with a computing setup that taps the unused power of its personal computers already in use around the corporation.

Now, instead of running simulations of new drug compounds on expensive high-end computers, company scientists can route the same computations to the company's maze of desktops and laptops for processing.

In setting up the grid, the $19- billion-a-year drugmaker is overcoming minor technical irritants and Excedrin-sized management headaches. But the potential payoff is huge. Bristol-Myers believes it can increase its research department's computing power fivefold immediately and as much as a hundredfold over the next few years, as more personal computers are tapped.

"It gives us a massive increase in compute power," says Richard Vissa, an executive technology director in Bristol-Myers' pharmaceutical research group. "And it opens the door for our scientists to look at new algorithms and new approaches that they might not have considered before, because there just wasn't enough compute power."

That can lead to viable new drugs. After all, Bristol-Myers is looking for any help it can get. Like other big-name pharmaceutical companies, the company's profit margins are being pressured by generic drugmakers. At the end of October, Bristol-Myers reported third-quarter net earnings of $314 million—75% lower than in the year-ago quarter. Beyond that, the company said it would restate earnings, after an accounting review showed it had pushed too much unneeded product onto its distributors.

Unlike other computing techniques that also juggle simultaneous computing tasks—such as massively parallel processing schemes that employ powerful servers—the goal of grid computing is to harness the underutilized processors in personal computers. Generally speaking, the assumption is that the processor in a typical corporate desktop or laptop machine is only in use 5% to 20% of the day. The processor usually sits idle, while employees go to meetings, tend to non-computer-related tasks, and, of course, go home at the end of the workday. Such downtime seems downright wasteful when one considers that today's PCs have the same processing power that supercomputers did 10 or 15 years ago.

At Bristol-Myers, however, every PC processor tied into the grid is employed full time. Platform Computing's ActiveCluster package sits on a server, and schedules and allocates work to the company's Windows-based PCs based on machine availability and capability. The package also includes agents—installed on all the PCs in the grid—that send PC status reports back to the server.

Bristol-Myers and Platform will say only that the pharmaceutical company has so far hooked "thousands" of PCs to its grid, declining to be more specific.

"We believe there is a significant competitive advantage to doing [grid computing]," Vissa says.

Gartner Inc., a computer market research group, earlier this year began to urge clients with sizeable numbers of desktop computers to "explore the potential of this promising distributed computing technology."

And organizations from southern California's Pacific Life Insurance (one of the largest insurance firms in the country) to Princess Margaret Hospital (a teaching hospital in Toronto) are building PC grids.

Bristol-Myers, according to industry watchers, is probably the biggest company to embrace grid computing. But as the company and its vendor set out to deploy grid computing, they found few guideposts.

"We're all learning," says Platform's Chief Technology Officer Songnian Zhou. Bristol-Myers confronted issues ranging from software conversions to security to the larger, administrative challenges of setting new PC management protocols and convincing people to "share" their personal machines.

"It's a complex management challenge," says Robert Batchelder, a research director at Gartner.

Overcoming Employee Resistance

Overcoming the resistance of the workers whose machines are connected to the grid was one of the biggest obstacles. Within Bristol-Myers, there was a group of "doubting Thomases," as Vissa calls them, who didn't believe grid computation could run on their PCs without disrupting normal desktop functions.

Vissa took to the road with a demonstration that proved grid computing would not degrade users' PCs in any way. Once people saw the demonstration, Vissa says, their comfort level rose and resistance declined.

And while some users were reluctant to embrace the grid, others couldn't wait to get their hands on it.

"Scientists always want more compute power so that they can try variations on various algorithms," Vissa says.

But with different groups vying for grid access, the company had to figure out a way to set work priorities. Bristol-Myers is handling that potential problem by establishing a governance committee of key scientists to schedule jobs.

There were other people issues. For instance, Bristol-Myers' facility managers—who are responsible for, among other things, keeping Bristol-Myers' utility bills down, would notice unattended PCs up and running. They'd leave a note reminding the owners to turn off their PCs at night—which, of course, would prevent the machine from receiving and processing work from the grid. And then there were the owners of grid-connected laptops who would take their machines home for a night or weekend.

"These are the things you have to wrestle with," Vissa says. He and his team now spend time educating people and bringing the company up to speed on grid-computing procedures.

Most of the technical challenges, on the other hand, were comparatively straightforward.

For instance, the scientific applications Bristol wanted to parse out to its Windows PCs were written for the Unix and Linux operating systems environments. The company went out and brought in a set of software conversion tools that automatically tailor the Unix and Linux applications to run on the Windows PCs.

The company also had to make sure its entire grid-computing programs were secured. Bristol-Myers' grid, which connects PCs on one campus in Connecticut and two in New Jersey, runs over an internal network. All PC computations run in a portion of the Windows operating system, called a security sandbox, that's virtually inaccessible to users. Still, Vissa says, his team is constantly monitoring the system to protect against any intrusions.

"Anybody that's in [information technology] that's not worried about security is not cognizant of the risks," he says.

Neither the drugmaker nor its grid software vendor would say how much Bristol-Myers spent on the project. Platform Computing prices ActiveCluster at $15,000 for the management software and $99 per PC. Those costs would be in line with a Gartner estimate that large grid-computing projects cost between $250,000 and $750,000 for software and services—or about the price of a set of new servers. Yet, the grid platform can expand exponentially by simply adding more PCs. Bristol's goal is to tie just about every PC in the company to the grid.

But while every PC might be a grid-computing candidate, not every computer application is. While perfect for big scientific or mathematical problems that can be easily parsed, computed and compiled, the resources required to run business applications that need to talk with corporate databases aren't practical on a grid. That's primarily because of the huge network traffic jams that would be caused by all the two-way communication between the processor and the data repository. For instance, Bristol-Myers has no plans to frame out a grid-computing scheme for its drug developers because such work demands access to the huge clinical-trial database.

Right now, Gartner's Batchelder says, grid computing is only really suited for very compute-intensive life-science, financial, and mechanical and engineering applications.

"It's not a wonder drug," he says.

What You Should Do When Considering Grid Computing

Do demos. Overcome resistance by showing workers that tying their PCs into the grid won't disrupt their jobs. Put things in order. Figure out which grid programs are going to run when. User-governance committees can help set priorities. Educate everyone. Bring the entire organization up to speed on do's and don'ts, like leaving grid-connected PCs on at night. Make conversions. If you're looking to partition pieces of a Unix or Linux program over a grid of Windows PCs, you'll need to tweak the application to run on Windows.

Getting More Out Of Computers

Computing tasks can be performed by any number of processing schemes, some of which fall under the loose heading of grid computing.

Symmetric Multiprocessing (SMP) A server computing system where two or more processors are managed by a single operating system. The processors share the same memory and input/output mechanisms.

Massively Parallel Processing (MPP) A server computing system where a hundred or more processors work on different parts of a task. Each processor has its own operating system and memory.

Grid Computing A scheme to take advantage of underused processors on corporate desktop and laptop computers. Tasks are broken down and assigned to individual processors by a master scheduler, much like MPP.

Utility Computing Also known as a hosted grid. A company activates and deactivates resources as needed from a large information-systems facility. The customer is charged for the services it uses. It's a modern version of computer time-sharing.

Sources: Giga Information Group, Aberdeen Group

Copyright © 2004 Ziff Davis Media Inc. All Rights Reserved. Originally appearing in Baseline.