NEW YORK, Sept. 13 - Pharma isn't known for its "power to the people" philosophy, but when it comes to computational power, pharmaceutical companies are embracing the idea.
Pharma is gradually adopting various forms of grid computing, and recent deals for this technology have given grid-software vendors a dose of optimism about the market potential for their tools. It also gives them a reality check about the capabilities and limitations of the technology for life-science informatics.
Both Novartis and Bristol-Myers Squibb recently signed on to use desktop grid solutions from United Devices and Platform Computing, respectively, while Avaki said in late August that Pfizer had acquired "components" of its data-grid technology. Meanwhile, GlaxoSmithKline, Merck, and AstraZeneca are active partners in the UK-based MyGrid project.
While "every deal is exciting," according to Yury Rozenman, director of life-science business development at Platform Computing, getting big pharma to sign on the dotted line indicates "the beginning of this market opening up."
But getting to this point hasn't been easy for providers of grid solutions, which harness the unused processing cycles of all computers in a network. Novartis and BMS, for example, evaluated solutions from several different vendors for nearly a year before signing on with UD and Platform. They still had to overcome a number of technical and cultural issues along the way.
Novartis, which is still negotiating the details of its agreement with UD, plans to implement the company's MetaProcessor agent on the desktops of up to 1,400 research staffers by the end of the year, according to Pascal Afflard, head of advanced computing and the Global Datacenter at Novartis. Afflard expects to extend that to 2,500 research desktops, and eventually beyond the research group into other corporate departments. "It's just a matter of getting the agreements," he said.
But testing and implementing the technology is "not trivial," according to Manuel Peitsch, head of research at Novartis. "Most new projects consist of about 40 to 50 percent technical questions and 50 to 60 percent process questions, such as cultural aspects and work aspects," said Peitsch.
The UD pilot had its share of both: There was a bug caught during the implementation, which Afflard said was addressed quickly by the UD team. Also, the technology brings with it the distinctive challenge of convincing desktop users that their computers won't be adversely affected by the software. After all, "PC means personal computer," noted Afflard, "so people had many questions."
Rich Vissa, executive director of global core technologies and informatics at BMS, which plans to install Platform's ActiveCluster on "several thousand" desktops, had a similar experience. "Grid computing could be the whipping boy of any kind of problem that someone has on their desktop. ...," he said. "We got a note from one user that said, 'I've been having an e-mail problem; do you know if I have the grid program running on my computer?' And we said, 'No, you've never had the program running on your computer,'" Vissa recalled.
Another unexpected obstacle came in the form of facility management policies: BMS and Novartis had to circumvent corporate requirements to turn off all computers at night in order to save electricity--to many a laudably frugal practice, but a showstopper for desktop grids, which benefit the most from computers that are on but not running other applications. Aside from overcoming the official mandate, "it's a cultural thing to get people to turn off their display but not their computer at night," Vissa said.
Finding the right applications to run on the architecture was also a big part of the evaluation process. "Some applications will not lend themselves to the technology," said Vissa, who cited as one consideration the lower memory footprint of desktop machines compared to Linux nodes.
While high-throughput docking was an obvious first choice for BMS and Novartis, other desktop-enabled applications are planned for both firms. In BMS' case, the desktop grid will free up cycles on several SGI machines that are currently running the docking programs. Afflard said that Novartis' desktop grid wouldn't take over specific cycles from another system, but would serve as one more option for the company's high-performance computing users.
Although the evaluation process of companies like Novartis and BMS eventually paid off in the form of sales for UD and Platform, other vendors are anxious to see more sales in less time. "The buying cycle is longer than we'd like to see," said Mark Weitner, vice president of sales and marketing at Parabon Computing.
While noting that the company has seen "an uptick in renewed interest" in its technology recently, Weitner said he's not seeing a speedup in deals just yet. Martin Stuart, vice president of life sciences at Entropia, agreed: "We're still at the stage where people want to see proof of concept."
Aside from the technical and cultural considerations that Novartis and BMS had to work through, other issues come into play when marketing grid computing to new prospects. For one thing, customers need to be sure that their chosen vendor has staying power--one factor that Vissa said BMS took into consideration when choosing Platform over its competitors.
In addition, Stuart said, vendors and buyers need to reconsider the licensing models for any third-party commercial packages they plan to run on distributed systems. Both per-seat and site licenses must be renegotiated. While Stuart said most software vendors are cooperative in this process, he noted it's an aspect of grid computing that potential buyers often overlook.
Another issue many vendors fail to consider is PC upgrade patterns at large organizations. Novartis' Afflard noted that the installation of the UD system on the company's computers is currently stalled because the entire research department is currently replacing its PCs. Stuart added that Entropia's system, which runs on Windows 2000 or later, is simply not compatible with many pharma desktops--a surprising number of which are still running Windows 95.
Because large enterprises often wait for the bugs to be ironed out of new systems, Stuart said he's only seen the transition from Windows 95 or 98 to 2000 or XP in many larger companies within the last year.
A final obstacle that Stuart pointed out is of the self-inflicted variety: Grid, distributed, peer-to-peer, and other similar incarnations have become victims of their own hype. Increasing media coverage of these technologies has led to confusion in the marketplace, Stuart posited, "and when a prospect becomes confused, the easiest thing is not to do anything."
However, he added, there is a bright side to the publicity deluge. Citing the Gartner Group's annual "Hype Cycle of Emerging Technologies" report, which tracks new methods from the initial "tech trigger" period through the "peak of inflated expectations," the "trough of disillusionment," the "slope of enlightenment," and onto the final "plateau of productivity," Stuart noted that desktop grid computing might be working its way from the trough to the slope phase right now, largely because users are discovering which applications work best with the architecture.
"People will only adopt a new technology when they have the opportunity to solve their problems, and the application space has been critical in that respect," he said.
Some observers say that vendors focused on desktop grid solutions are missing out on larger opportunities in the life sciences market. According to Philip Werner, vice president of product management for Avaki, "the feedback we're getting is that the cost benefit [ratio of PC desktop grids] is not all that high."
Administrative overhead costs can be considerable, he said, and another hidden cost is the tendency of users to "mess with" or turn off software on their desktops. Avaki, which provides both compute grid and data grid technology, addresses the compute problems of networked enterprises by solving the data integration problem first, he said. While agreeing that interest in implementing grid approaches is mounting, Werner said the driver isn't the technology so much as in "understanding that they have a distributed system problem."
Afflard and Vissa both noted that their desktop grid installations comprise only a small piece of what is likely to be a much larger grid-based strategy. Afflard said that Novartis is talking with both Avaki and Platform about additional "technology surrounding the grid concept," while Vissa said BMS is considering plugging its desktop cluster into a larger Platform system that includes its Linux and Unix servers.
But the benefit of harnessing the compute cycles that would otherwise be wasted on its PCs was enough of a selling point to go with the desktop solution, according to BMS and Novartis. Said Vissa, "When you look at the leverage you get with thousands of PCs versus maybe a hundred or so Intel nodes in a Linux cluster, your first focus in this space is going to be your desktops. That's where you're going to get the biggest bang for your buck."
Afflard noted, however, that it's important to keep the limitations of desktop PCs in mind when considering the approach: "They can be used effectively for high-performance computing, but their predominant use will always be [Microsoft] Office."
This article originally appeared in BioInform, GenomeWeb's sister publication.