Orion Multisystems is eyeing the bioinformatics market as a key early adopter for a new “cluster workstation” supercomputer it will begin shipping in October.
Colin Hunter, CEO of the Santa Clara, Calif.-based startup, told BioInform that “biosciences will be very important initially” as it rolls out the DT-12 workstation. The product packs 12 processors into a single desktop box that delivers 18 Gflops of performance, consumes less than 220 watts, and is priced under $10,000.
Orion has partnered with The BioTeam to include the consulting firm’s iNquiry software package in a configuration called the Orion Cluster Workstation for Bioinformatics. The company is also working with another bioinformatics consulting company, Callident, to include that firm’s BioBrew application suite on the DT-12, and has an additional collaboration with Wolfram Research to pre-install Wolfram’s gridMathematica scientific software on the workstation.
Hunter said that he and Orion co-founder Ed Kelly became “fans of workstations” while employed at Sun Microsystems in the 1980s. The two later went on to found chipmaker Transmeta, where Hunter said they “learned the importance of performance per watt” in an effort to drive development of smaller, lighter, and longer-lasting laptops. “But based on our backgrounds at Sun, we became convinced that this technology could be used to make workstations that were much more high performance, and the reason was the advent of cluster technology in high-performance computing,” he said.
Orion has spent the last two years developing a system that delivers the parallel performance of a Linux cluster in the form of a desktop workstation — a once-ubiquitous tool in the scientific and technical computing market that has waned in popularity with the advent of low-cost commodity clusters. But Orion is convinced that the demand for workstations is still strong in the sci/tech market — provided it’s a workstation that can deliver the power of a supercomputer.
The Cluster Workstation
The company’s cluster workstation architecture is based on the same concepts underlying mobile computing and blade servers. For the DT-12, 12 low-power chips (1.5 GHz Efficeons from Transmeta) are connected via gigabit Ethernet on a single circuit board, along with the switch. “Because it’s all on the same board, we can run the gigabit Ethernet at very low power,” Hunter said. “Instead of two watts per node, it’s more like 125 milliwatts per node.”
The company is also developing a larger, deskside system called the DS-96 that will contain 96 processors, deliver 300 Gflops of peak performance, and consume less than 1,500 watts. Orion plans to launch the DS-96 in late Q4, and will price it at less than $100,000.
Orion is initially targeting market sectors where Linux clusters are prevalent, making the cluster-friendly bioinformatics community a fairly obvious place to start. The company has built its management team with this goal in mind, hiring Juli Nash Moultray — formerly biology market manager for SGI — as industry marketing manager, and Stu Jackson — formerly director of bioinformatics at Incyte — as applications engineer.
Jackson, a long-time fan of Linux clusters who built Incyte’s 1,200-CPU system himself, said that the cluster workstation offers obvious advantages for power users. “At one point I had a 20-CPU cluster running under my cubicle,” he said. “That particular cluster probably drew around 1,200 watts of power and was actually less powerful … than the Orion machine that I’m using right now that draws less than 200 watts.”
Orion expects the system to appeal to developers writing code for clusters. These developers “need a cluster to write their programs, but ... they’re always writing programs that break, and everything goes down and the people who want to use the cluster to get their job done can’t,” Hunter said. The workstation will enable bioinformatics teams to move those development tasks to a dedicated system, freeing up the large machine for end-users.
Hunter said the cluster workstation also offers advantages for end users frustrated with overtapped shared resources. “If the machine runs the code three times faster on a giant system, but you’ve got to wait for hours to get your time on it, you might as well just run it on your local workstation, because you’ll have 100 percent of your time being used for your job,” he said.
Performance … and Competition
Hunter said that, like most clusters, the system scales with the number of nodes. Applications like Blast, therefore, also scale linearly, while those that require more inter-processor communication, like molecular dynamics applications, don’t scale as well. Jackson said that the company is putting together Blast benchmarks for the system, but they haven’t been published yet.
Orion may be first to market with a desktop-sized supercomputer, but it’s not the first to offer an off-the-shelf bioinformatics cluster. Companies like Time Logic and Paracel have been selling dedicated “accelerated” bioinformatics systems for years, and more recently, large IT firms like Sun, Apple, and IBM have begun courting the life science market with pre-configured bioinformatics boxes that can fit under a researcher’s desk or lab bench [BioInform 08-09-04].
Hunter said that Orion has an edge over these systems because it offers more computational power in a smaller footprint with lower power consumption. “They’re basically putting 1U devices in a short rack with wheels on the bottom, and the problem is that all the power difficulties associated with 1U systems are still there,” he said. “It’s just that you have fewer of them.”
Hunter added that competition for off-the-shelf bioinformatics boxes is ultimately a sign that Orion has entered the market at the right time. “It does validate that there is a demand there,” he said. “Customers are saying, ‘Look, I like clusters. I just want to have my own personal one.’”