Beset by corporate-level financial woes in recent years, Sun Microsystems hasn't had the resources to maintain a strong presence in the life science market, but that's about to change, according to the company's new life science management team.
Next week at the Life Sciences Conference and Expo in Boston, Sun will be showcasing a new product targeted to the bioinformatics market. Dubbed the Discovery Cluster, the system is based on the company's Opteron-based "Galaxy" servers, and is priced at under $94,000 for a single rack. Three racks give a teraflop of performance for less than $282,000.
The system provides high-end computational power at a price far lower than Sun has ever offered before, Joerg Schwarz, director of healthcare and life sciences at Sun, told BioInform. "That's the message that we want to send out to the community — that we're back, we're back with a vengeance, and we do offer unprecedented price/performance, which is not necessarily what Sun has been known for in the past."
Last July, Sun tapped Schwarz to rebuild its business development and sales efforts in healthcare and life sciences. Previously, he led the company's global business development efforts in academic high-performance computing. Stefan Unger, who was previously in charge of business development for computational biology under Sun's global education and research group, has joined Schwarz's team as business development manager for life sciences.
"That's the message that we want to send out to the community — that we're back, we're back with a vengeance, and we do offer unprecedented price/performance, which is not necessarily what Sun has been known for in the past."
The centerpiece for the company's revamped life science strategy is the Discovery Cluster, a pre-assembled rack based on the Sun Fire X2100 64-bit x64 server, which is powered by the AMD Opteron dual-core processor. According to Sun, the X2100 delivers up to one-and-a-half times the performance, and uses about one-third of the power of competing systems from IBM, HP, and Dell, but costs a fraction of their price.
According to data provided by Sun, in 16 out of 19 Blast benchmark cases, the X2100, which has a list price of $745 per server, outperformed a Xeon-based Dell Precision Workstation, which costs around $1,400 per server.
The cluster supports the Solaris 10 operating system as well as Red Hat Linux, Suse Linux, and Windows.
While it comes with a suite of bioinformatics freeware that users can install, the Discovery Cluster isn't intended to be a preconfigured bioinformatics box, Schwarz said. Rather, it's a hardware system designed to provide the right level of compute power for a range of life science applications at the lowest possible price.
Unger said that the Opteron chips are particularly well suited to life sciences computing. "We like the Opteron because it has nicely balanced performance," he said. "It has very high integer performance, which you use for all the bioinformatics stuff, and it also has very good floating point performance, which people would use for molecular modeling, docking studies, and simulation."
Unger said that the initiative for developing the Discovery Cluster grew out of the so-called Galaxy line of servers that Sun began selling in the fall to regain its footing in the server market. "It was a fresh start," he said.
By using Opterons instead of its traditional SPARC chips, Sun was able to lower the prices of the Galaxy line to compete with Dell and others selling servers built with commodity processors.
For the Discovery Cluster, Sun reduced the price even further by including only those features necessary for its target user base — experienced bioinformatics customers.
"What our customers want is bang for the buck," Unger said. The X2100 server at the cluster's core "does not have a lot of extra bells and whistles in it because the people who are doing discovery don't want that. They don't want CDs in every node, and they don't want necessarily a huge amount of storage in every node," he said.
Schwarz said that the dual-core CPU in the X2100 also helps reduce the cost. "Normally people use two single-core CPUs, and that takes more power, it takes more space, and costs a lot more because the pricing is different."
In addition, he said, "we also decided not to spend too much money on, for example, InfiniBand interconnect because we found that gigabit Ethernet is perfectly fine for the purposes of discovery."
While the one-size-fits-all approach to the cluster is what enables the low price, Schwarz said that Sun will customize it to include higher-end servers, more memory, or lower-latency interconnect for those customers who request those features.
The Discovery Cluster is a departure from Sun's last offering for the life science market, a four-node system it launched two years ago called the Sun Fire Starter Cluster for Bioinformatics, which was priced under $22,000.
"This is definitely targeted at a more mature group that knows what they're doing," Unger said, "and that's part of the reason why we didn't put this thing in 50 different versions and tie it up with particular software, because we think the more experienced user knows what they want, they know their software, and they're going to end up customizing this thing anyway."
— Bernadette Toner ([email protected])
ABI Tests Sun's Web-Based Utility for Life Science Computing
The Discovery Cluster isn't the only offering in Sun's revamped life science lineup. The company has also launched a new service that provides online access to its hosted computational services for $1 per CPU/hour.
Applied Biosystems was a beta customer for the on-demand service, called the Sun Grid Compute Utility (available at http://www.network.com/).
Francisco De La Vega, senior director of computational genetics at ABI, told BioInform that the company was faced with a flood of SNP data in the fall when the international HapMap project released a batch of verified SNPs and the company had to develop millions of new TaqMan SNP genotyping assays as quickly as possible.
De La Vega said that the most compute-intensive part of the design process — ensuring that the assays are specific in the genome — would have taken around three months on the company's internal compute farm. "So our choice was either take the hit in terms of arriving late to the marketplace," or invest in new in-house hardware, "which didn't make sense in terms of cost effectiveness because this was a peak job."
After hearing that Sun was planning on launching a utility grid service, De La Vega said he was attracted by the low price and opted to join the beta program. The results, he said, were "better than expected." While he estimated that it would take around 35,000 CPU hours to complete the project, it actually took 26,000 CPU hours and less than a week, he said.
The only bottleneck in the process was porting ABI's code from its Linux/Intel architecture to Sun's Solaris/Opteron architecture, which took several weeks, De La Vega said. Nevertheless, he added, he would use the platform again for similar peak jobs in the future.
Joerg Schwarz, director of healthcare and life sciences at Sun, said that the Discovery Cluster and the utility grid are at the two ends of a "continuum" of offerings that the company is putting together for the life science market. This ranges from the ability to "own and operate your own clusters … to offers where we will build a cluster on the customer's premises and operate it. We even offer to turn it into a utility, so customers can pay us per use," he said.
Customers can also arrange for Sun to build and host specific hardware configurations that they can pay for on a usage basis.
Finally, the utility grid offers a quick and easy way to access additional compute power at any time, with no startup costs. "All people need is a PayPal account," Schwarz said.