Linux Clusters Rev Up on PC Blades
Energy, Medicine Benefit from Linux Clusters
When it comes to Linux clusters, mainframes have long drawn the largest acclaim. Now, though, scientific and technical imaging specialists such as Paradigm and Electro-Optical Sciences (EOS) are praising the gains they're seeing from Linux clusters on PC servers, even without blades.
According to Mark Melenovsky, an IDC analyst, these two IBM customers represent a couple of industry trends particularly visible over the past year: a move to Linux-based PC servers, and the rise of blade servers.
IDC's latest quarterly numbers show that the overall server market remained flat from Q2 2002 to Q2 2003 at about $11.7 billion, Melenovsky says. Meanwhile, however, spending on Linux servers shot to about $650 million, a year-on-year gain of 39.5 percent, according to IDC. These statistics don't even include mainframes running Linux on top of OS390 and zOS. "We're basing our numbers on the primary OS only," he says.
Growth in the blade server market was even stronger..Out of about 1.2 million servers shipped in Q2 2003, blade servers accounted for 40,000 units, a huge increase over the 7,000 to 8,000 units shipped in Q2 2002.
"Blade servers are still a nascent market, but they're finding a very strong niche in the Linux cluster and high performance computing market," Melenovsky maintains.
For its part, Paradigm recently installed several hundred IBM BladeCenter nodes, all running Linux, within its own data centers in Houston, Texas; Moscow, Russia; Woking, UK, and Mumbai, India. BladeCenter uses Intel's 3.08 GHz Xeon processor. Meanwhile, Paradigm also sells applications for Linux clusters to some of its customers in the oil exploration industry, notes Philip Neri, Paradigm's solutions marketing director.
Paradigm produces software products such as Earth Domain Imaging, for modeling the earth's interior; Focus processing software; and GeoDepth, for modeling velocity and depth, for instance.
Paradigm's customers do their own hosting, according to Neri. "They typically want to retain a high level of confidentiality."
EOS, on the other hand, recently ported a Red Hat Linux-based cancer diagnosis application from homegrown PC clusters to bigger and more powerful eServer clusters, running at IBM's recently launched "Deep Computing on demand" facility in Poughkeepsie, NY.
The current architecture at the Poughkeepsie facility consists of Intel-based IBM eServer x335 nodes, components of IBM's eServer Cluster 1350. Over time, IBM expects to add blade and AMD technologies. Aside from scientific and technical fields, IBM is targeting its Linux cluster services at the financial market, along with entertainment disciplines such as animation.
Users and analysts point to huge performance boosts from the use of eServer Clusters, plus cost savings and other advantages associated with running open source Linux in place of "proprietary" Unix.
"Many computing-intensive applications that could previously only be done on large proprietary Unix systems are now being run on Linux," observes IDC's Melenovsky. Some of these apps are still being operated on rackmount systems such as Dell's Rack-Optimized Server.
"There are also lots of smaller vendors in the 'rack pizza' space," the analyst adds. Certain smaller competitors, such as Rackable and RackSaver, have been working their way into the blade market, too.
"Some of our customers are still running our oil exploration software on proprietary Unix systems," acknowledges EOS's Neri.
At this point, though, there aren't that many Unix-based supercomputers still on the market, Neri contends. "Convex is gone," he observes. "Cray was bought by SGI. SGI resold Cray machines, before selling off its Cray business and getting into the market themselves. There's also IBM's AIX p690 system."
Paradigm first started running Linux clusters internally back in 2001. "The blade architecture is relatively new. When we took a look at it, though, we saw that we'd be getting vast gains in floor space, along with greater energy efficiency," according to Neri.
EOS initially ran Linux clusters on boxes put together inhouse. "The parts are easy to get," says Alex Bogdan, a principal developer at EOS.
The company's move to outsource hosting to IBM stemmed from needs for faster performance and higher scalability. The medical imaging house is now pursuing FDA certification for MelaFind, an application designed for noninvasive early diagnosis of potentially fatal melanoma, a form of skin cancer.
"We've already achieved (FDA) certification for another application, called Difoti," says Martin Frusin, EOS's director of business development. Dentists use Difoti for early detection of tooth cavities.
"Our plans now call for extending our 'platonic computer vision' technology into various types of epithelial or 'tissue' cancers, such as skin and oral cancer," Frusin adds.
MelaFind will utilize advanced search and neural networking technologies, running on Linux clusters, to help medical pros decide whether surgical biopsies are needed. Other components include a computerized handheld scanning device, which doctors will use to take digital images of suspicious lesions on patients' skin, as well as a database for storing these images and related records.
According to Bodgan, it took EOS only a couple of hours to port 10 or 12 miniclusters from MelaFind to IBM's supercomputing architecture, and to optimize the existing app for IBM's Parallel Virtual Machine (PVM). Performance gains then showed up almost instantly. "Jobs that once took about a month to complete can now be run in about 10 to 12 hours," he says.
EOS's database, however, is not yet large enough to require IBM's Universal Database (UDB), Bogdan says. "We're still using MySQL, and it's doing just fine for us right now."