San Francisco – At Caprion Pharmaceuticals Inc., a drug discovery firm in Montreal, the name of the game is pain relief. The company uses some hefty proteomic research and high-tech instruments to scrutinize tissue samples and, in time, find new treatments and remedies.
The company employs a complicated method of drug discovery, comparing diseased human tissue with healthy samples through mass spectrometers – devices that reduce molecules to their components. The in-depth comparison gives scientists greater knowledge of ailments and brings the pharmaceutical firm that much closer to finding cures or “at least a way to reduce the effects” of diseases, said Bernard Gagnon, Caprion’s director of IT.
If Caprion’s scientific work offers insight into the world of medicine, its computing infrastructure offers insight into world of IT. Thanks to a mindset that suggests “two heads are better than one,” this drug company is at the forefront of a trend that could change the way enterprises consider network infrastructure.
On the surface Caprion’s work deals with medicine, but the company requires solid computing to support its biochemical know-how. After all, each tissue sample Caprion works with generates 100GB of data, Gagnon said, adding that the firm produces a terabyte of information each month. The company uses complex algorithms to turn raw data into usable information and, ultimately, new drugs.
Big number-crunching jobs require big computers to get the work done. Gagnon said the company did consider purchasing a supercomputer – perhaps something like Cray Inc.’s MTA system – but Caprion didn’t care for the price of entry. Supercomputers are expensive, often because they ship with the latest, fastest technology on board – the best hardware manufacturers can offer with a price tag to suit.
Caprion instead sought a way to connect numerous smaller computers and create a distributed supercomputer, thereby avoiding the massive up-front cost associated with Cray’s work.
Caprion purchased Sun Microsystems Inc.’s Grid Engine, a program that combines computing power from numerous boxes, turning a company’s loosely-defined collection of CPUs into an I/O juggernaut.
Gagnon said Caprion connected a Sun Solaris box running the Grid Engine software to eight eight-way computers to create the high-tech net. Compared with the supercomputer option, “we cut the price by more than half,” he said.
Grid computing is growing popular. Sun, Hewlett-Packard Co. and IBM Corp. offer grid programs that connect a number of computers together for simple processor sharing and quicker job processing.
Caprion isn’t the only company employing grids, either. Gagnon said many life-sciences firms use the technology for large mathematical problems. Outside of this sector, “distributed computing” projects bring together numerous computers and borrow unused clock cycles for massive problems. Witness the SETI@Home project, which links boxes around the world to decipher signals from outer space.
But grid computing isn’t just about sharing clock cycles and quickstep discoveries. It’s also about networking. After all, wires stand between the grid’s individual elements and comprise the veins connecting the net’s extremities.
As such, grid computing could have a profound effect on network architecture, particularly the types of wires companies use to bring devices together, said Howard Asher, Sun’s group director, global life sciences group.
Asher said grid computing is a catalyst for fat conduits between network elements, explaining how this technology would increase the use of gigabit Ethernet and multimode fibre-optic cables in the enterprise.
“We have to open the pipes and get that throughput,” he said during an interview at Sun’s SunNetwork conference in San Francisco. “That’s why Cisco (Systems Inc.) is such an important partner for us.”
Asher also suggested that grid computing would result in wireless advancements. He pointed to the California Institute for Telecommunications and Information Technology (Cal-IT) wireless research lab as an example, where Larry Smarr and a team of scientists are working on secure, high-speed wireless networks for extraterrestrial exploration. It’s no stretch to port Smarr’s work into the grid space, where “wireless high-speed grids” would result, Asher said.
He noted that bandwidth is important in grid environments. Imagine a CAD operator waiting for his computer to render an image, Asher posited. This long-term task, which might take 24 hours if the file is large enough, would take just minutes with grid access.
Now imagine that the operator is trying to access those extra clock cycles while other traffic is traversing the LAN. If the connecting pipes are too small, the designer’s project could impede other data travelling the network and vice-versa.
Asher said grid computing would result in companies seeking more bandwidth for connected elements. But as much as Asher believes in bigger pipes, Gagnon from Caprion said he believes in planning. The company implemented its grid only after testing I/O speeds and making sure it had the requisite infrastructure for fast data transfers.
Caprion built a separate network consisting of gigabit Ethernet service specifically for its eight number crunchers. “Doing it any other way would have caused problems for our regular LAN traffic,” Gagnon said.
He also said Caprion doesn’t need more than the 64 processors it uses in the grid -yet – but John Tollefsrud, Sun’s grid marketing manager, said much larger grid projects are coming down the pipe. Sun plans to be ready for the ensuing CPU-networked age with a global or super wide-area grid engine in the near future.
Aside from proper planning, Gagnon said companies considering grid computing should also employ a modicum of skepticism; don’t believe the hype, particularly concerning the Linux OS and its supposed superiority in grid environments.
“There’s been a lot of press about Linux…clusters being a lot cheaper. We did a study and found that’s not so,” Gagnon said, explaining that hardware, rather than software, is important when it comes to price. “As long as you decide on the size of the box, (the fact that the box is) Linux, UNIX or Solaris doesn’t affect price at all.”
Caprion uses Solaris in its grid-connected boxes, Gagnon said.