NEC has recently sharpened its product marketing strategy for the life sciences community: Last week, the Japanese company’s US arm, NEC Solutions, said it plans to launch a blade server based on the 64-bit Itanium 2 processor in September, and that it has targeted bioinformatics as a primary application area for the system.
The bioinformatics community has embraced the higher density and lower power consumption of slimmed-down blade servers, so even though NEC currently claims minimal market share in the US life science IT sector, the company is counting on this field as a core customer base for its upcoming offering.
NEC is best known for its vector-based 35.6-Tflop Earth Simulator housed at Japan’s Marine Science and Technology Center in Yokohama. The company launched a Bioinformatics Business Promotion Department in 2000, which offers a range of targeted software and hardware solutions for the Japanese market, but NEC has lagged a bit in the US life sciences sector.
The company has recently taken some steps to firm up its strategy in this area, however, according to Beth Makosey, director of corporate communications for NEC Solutions. She said that the firm formally launched a healthcare solutions division around nine months ago, and recently hired Lauryn Jones as director as director of the new group. The blade server, Makosey said, is expected to play a “big part” in the company’s future healthcare strategy. Jones was not immediately available for comment.
NEC joins another Japan-based IT firm, Fujitsu, which is also ramping up its strategy for entering the US life science market. [BioInform 05-03-04]. But while Fujitsu’s strategy is based on its 1,920-processor 2.5-Tflop BioServer, NEC is hawking a smaller-scale hardware offering. Its Express5800/1020Ba blade server will offer two 1.6 GHz Itanium 2 processors per blade. Nine blades fit into a 19-inch, rack-mountable 10U chassis to deliver around 100 Gflops of processing power per chassis at a starting price of $79,000.
Blade servers offer a smaller footprint, lower power consumption, and simplified cabling compared to standard 1U rack-mounted servers — advantages that the life science research community has been quick to exploit. RLX Technologies, one of the first blade providers, scored an early win with a 700-blade cluster at the Sanger Institute in 2002, and other vendors such as IBM and HP have found that blades are an easy sell for most bioinformatics groups.
In March [BioInform 03-29-04], the University at Buffalo’s Center of Excellence in Bioinformatics chose an IBM BladeCenter cluster over several other available hardware options to augment a 4,000-processor Dell cluster it installed in 2002. In addition, Australia’s Commonwealth Scientific and Industrial Research Organization’s Bioinformatics Facility has installed a 66-blade system from Dell; the Hartwell Center for Bioinformatics at St. Jude Children’s Hospital has installed a 280-processor IBM BladeCenter; and RLX said that it has recently installed a 500-blade system for biological research at Los Alamos National Laboratory, as well as “a few” blade systems in some pharmaceutical firms, whose names it is not able to disclose.
NEC claims that its upcoming offering will have some advantages over the competition, however. Scott Schweitzer, product manager for NEC’s Itanium server family, said that most blades on the market today offer 4-6 GB of memory capacity, but the NEC system supports up to 24 GB of memory with currently available 2 GB memory chips. In addition, he said, the system offers improved networking capability via a 4x InfiniBand adapter that can deliver up to eight simultaneous communications paths over a single cable.
The Express5800/1020Ba Blade server is a third-generation blade system for NEC, Schweitzer said, but is the first the company is launching in the US, and its first 64-bit blade. NEC will be the first firm to market with an Itanium blade, Schweitzer said, because the company was able to overcome some challenges associated with the considerable amount of heat the chips produce. Each chassis consumes around 5 kilowatts of electricity, Schweitzer said.
But some observers question whether an Itanium chip is really necessary for most bioinformatics applications, especially Blast. Scott Farrand, vice president of systems engineering at RLX, said that his firm has no immediate plans to move to Itanium. “We’re not seeing a landslide of customers asking for that,” he said. For customers looking for a low-cost Blast farm, he said, most find that a Pentium 3 is sufficient, although “increasingly we find people moving more to the dual Pentium Xeon because there’s better price performance, and it’s also good for protein folding applications.” Blast is an integer-based application, Farrand noted, “and almost any reasonable CPU today can do integer arithmetic very quickly.”
Jeff Skolnick, director of Buffalo’s Center of Excellence in Bioinformatics, agreed that the choice of chip architecture often depends on the application. “There are certainly applications where 64 bits are absolutely essential,” he said, “But for Blast or sequence analysis, our experience is that the extra precision doesn’t really make a difference.”
NEC said that it has developed a math library for the Express5800/ 1020Ba Blade server that includes life science applications such as Blast, Gaussian, Charmm, and Amber, but the company did not provide performance details before BioInform went to press.
Dennis Lam, a project manager in NEC’s Advanced Technical Computing Center in Houston, confirmed that the company has optimized Blast to run on the blade server. “We have focused a lot of our efforts to tune it and make sure it runs very well on the blade,” he said.
Express5800/1020Ba Blade Server Features
- Dual 1.6 GHz Itanium 2 processors (12.8 Gflop per blade)
- 12 dual in line memory modules for 24 GB of memory (2 GB/DIMM)
- 4x InfiniBand PCI-X host adapter with 1 GB/s throughput
- 9 blades, with 18 processors, per 10U chassis
- Two on-board 73 GB U320 SCSI hard drives
- Management interface controls all blades in the chassis and other blades connected over IP
- $79,000 for chassis, nine two-way blades (each with 2 GB of memory), 136 GB hard drive storage