Microsoft plans to launch a special version of the Windows operating system next year for high-performance computing, and a company executive told BioInform last week that the life science sector will play a large role in its marketing strategy for the new product.
“The life sciences are extremely important in terms of what we’re doing,” said Greg Rankich, senior product manager at Microsoft. “There are three markets that we identified as fast growing, that fit our scenario, that we are targeting towards. Those include life sciences, manufacturing, and the energy industry.”
Last week, Microsoft demonstrated a beta version of the system, called Windows Server 2003 Compute Cluster Edition, at the Supercomputing 2004 conference in Pittsburgh. Microsoft tapped discovery informatics firm Optive Research as its partner for the demo, along with HP and AMD.
The companies ran Optive’s Concord program for 3D chemical structure generation in the demo. Optive has been benchmarking the Concord software as part of a partnership with Microsoft that began earlier this year [BioInform 02-23-04].
Bryan Koontz, CEO of Optive, said that the company tested Windows against three Linux distributions — Red Hat Enterprise Linux 3.0, Red Hat 7.3, and SUSE 9 — and found that Windows achieved “platform parity on a performance basis.” The performance was “about par” with Red Hat 7.3, more than 50 percent faster than Red Hat Enterprise 3.0, and around 30 percent faster than SUSE 9, he said.
“The charge that we gave to our development team that was doing the benchmarks was, ‘Try to try and be objective when you’re running these tests.’ We did not spend a lot of time fine-tuning either operating system. We wanted to get as much of an idea of how it performed out of the box as possible.”
Koontz said that Optive was “skeptical” about the performance of Windows before running the benchmarks, and said the positive results were “really surprising to us.”
The experiments “prove to the user community that — at least when thinking about performance — Windows should at least be considered,” Koontz said. Now, he added, “it’s up to Microsoft in large part to get out there and start doing a good job marketing that platform, that operating system, to that particular segment.”
Market resistance to the Windows operating system is likely to be high at first. Koontz acknowledged that despite Windows’ performance, “there may be a whole bunch of other reasons why a company would end up going with Linux,” and that “Linux and Irix are still the platforms of choice for our customer base.”
Rankich ceded that converting Unix and Linux HPC customers to Windows “[is] my biggest challenge in the next year.” Nevertheless, he said, “it actually has been surprisingly easy and pleasant so far.” Rankich said that the HPC crowd doesn’t have the zealotry of some other computing communities when it comes to their choice of operating system. “For many of these people, at the end of the day, all they care about is performance. They want the fastest performing system, they want to make sure it’s going to work, and they want to make sure they’re going to get their research done,” he said. “There isn’t that religion about operating systems that you may find in other areas.”
He added that benchmarks like those generated like Optive will be a key part of Microsoft’s HPC strategy. “It’s my goal to not only be price competitive, but to also be performance competitive and be able to say, ‘Look, it’s not Microsoft trying to charge way too much for their product. We’re going to show you that the performance is there.’ And if we can do it cheaper than a Linux version, that says a lot.”
Rankich said that Microsoft is also working with other software vendors in the life science market, such as Accelrys, Tripos, and Turboworx, and is partnering with the biology group at the Cornell Theory Center to port publicly available bioinformatics algorithms like Blast and Hmmer to the new edition of Windows.
Microsoft’s goal, he said, is to have the “top 20” bioinformatics algorithms ported and tuned for the new system by the time it is released in mid-2005. This has posed a bit of a challenge for the company “because there’s such a huge group of [bioinformatics] users using open source,” he said. “If you go to the manufacturing or the oil industry, they’re pretty much all commercial applications. There are very few open source applications. So life science is a little more of a challenge to us in that sense.”
While Microsoft is known for its corporate-level opposition to open source software, Rankich said it’s not IP issues that pose a challenge in this case. Rather, the challenge is “more from just a partnership and engagement perspective.”
As opposed to partnering with a software vendor like Optive, “when you talk about the open source community, there are many owners,” he said. “So we said we’re going to use an academic [group] to port it, tune it, optimize it for Windows, and then we’ll push it back out to the community.”
The upcoming Compute Cluster Edition for Windows Server 2003 will include the operating system, middleware for messaging between compute nodes, and job schedulers and other cluster-management software tools. The initial version of the system will support “hybrid” 32-bit processors with 64-bit extensions like AMD’s Opteron and Intel’s EMT-64 architecture, with support for Intel’s Itanium 2 64-bit chip still “on the roadmap,” Rankich said. The beta system that Microsoft, Optive, and HP were demonstrating at Supercomputing 2004 ran on Opteron processors.
Optive entered into a partnership with HP last year [BioInform 10-23-04] to test its software on Itanium 2, “but we just didn’t see enough demand for [the processor] in discovery informatics,” Koontz said. “There may be other application areas where it’s doing just swell, but frankly, we just didn’t see the demand for it.”
Opterons, on the other hand, “are catching on with our user base,” he said.
Very few life science applications have been ported to the 64-bit architecture, which has stalled adoption of Itanium 2 in the sector, Koontz said. Hybrid chips like Opteron, however, enable customers to run their current applications “and as they do find opportunities to port things to 64-bit, the Opteron system can support a mix of 32 and 64, so they can get the little boost in performance if they want.”