“Customers in the bio space have been asking for clusters that accelerate bio codes — some will invest in FPGA and Itanium, but others want clusters — [so] hence, the demand [is there for SGI to meet].”
“Almost every other code used on a cluster would see some acceleration.”
SGI Partners With eXludus to Launch Accelerated ‘BioCluster’ Altix System
Premium
SGI this week released the BioCluster, a version of its Altix XE cluster platform bundled with capacity-management software from eXludus that SGI said provides a significant speedup for life science applications.
In benchmarks, a 64-processor configuration of the BioCluster demonstrated a 70 percent acceleration for HMMer and a 90 percent acceleration for Blastn compared to an Altix XE cluster without the eXludus Grid Optimizer software, SGI said.
The BioCluster was designed to outperform SGI’s Altix XE240 cluster, which includes PBS Pro cluster-management tools but not Grid Optimizer.
Deepak Thakkar, bioscience segment marketing manager at SGI, told BioInform that SGI has so far limited its benchmarking to HMMer and Blastn “because we are looking at what …codes people are using the most.” However, he said, “almost every other code used on a cluster would see some acceleration.”
He said that while the company has not yet signed on any users for the new system, a major academic institution in China has expressed an interest in it.
SGI has been planning to launch the BioCluster since at least May, when Thakkar told BioInform that SGI was planning to partner with eXludus on a bundled system targeted at life science customers [BioInform, May 4, 2007].
Grid Optimizer was designed to work in tandem with cluster-management tools like PBS Pro to improve cluster performance by optimizing workload scheduling. While the technology is applicable to all high-performance computing applications, eXludus has identified the life sciences as its first target market for the software.
In May, Stephen Perrenod, vice president of sales and marketing for eXludus, told BioInform that the company has seen the most interest in its technology from life science users because they run so many applications that work best in parallel, like Blast and HMMer.
Dale Geldart, executive vice president of sales and marketing at eXludus, told BioInform this week via e-mail that the partnership with SGI also “allows users to take full advantage of multicore processing resources.”
He added that the performance gains “are realized through a simple software installation with no other hardware or application changes required.”
Paul Lu, an associate professor in the department of computing science at the University of Alberta and an SGI customer, said that BioCluster’s “turnkey” approach could be of interest to potential customers.
“The product choices out there are immense. A lot of people don't have the expertise to pick and choose hardware and software,” Lu said.
Thakkar said the BioCluster is faster than the Altix ICE, a “green” computing platform it launched in June that consumes 87 percent less power than equivalent clusters [BioInform, June 29, 2007].
Thakkar said that BioCluster is a better option than Altix ICE for customers looking for speed. “There are time savings in terms of cycles you need to run,” he said.
He added that SGI plans to extend its bioinformatics offerings for ICE, which currently supports Blastn. “We are rapidly moving forward in making it a more complete appliance by extending its breadth,” he said. “We are extending it with Smith-Waterman, HMMer and ClustalW, all for genomics and proteomics.”
BioCluster isn’t SGI’s only accelerated bioinformatics offering. The company has an ongoing collaboration with Mitrionics, a field-programmable gate array programming firm, to develop an FPGA-accelerated implementation of Blast that runs on the SGI RASC architecture.
Earlier this month, FPGA provider Xilinx said that an SGI Altix with Xilinx Virtex-4 FPGAs ran Mitrionics’ Blastn 900 times faster than an equivalent cluster.
Xilinx said the Blastn benchmark required around three weeks to complete on a 68-node AMD Opteron cluster, but less than 33 minutes on a 64-processor Altix 4700 system with 35 RC100 RASC blades. Each RC100 includes two Xilinx Virtex-4 LX200 FPGAs.
Thakkar said that the Blast FPGA appliance is "gaining momentum," and noted that the two systems are expected to appeal to different users.