Its user-friendly desktops and PowerBooks are perennial favorites among bench biologists, but historically, Apple Computer hasn’t been a contender in the high-performance life science computing market. That perception is likely to change, however, after last week’s unveiling of the biannual Top 500 supercomputer ranking, where a cluster built with 1,100 Apple G5s at Virginia Tech debuted at the No. 3 spot (see story on p. 4 for further details on the Top 500 list).
“Apple is going to have a lot more visibility in the high-performance computing market,” predicted John Humphreys, senior research analyst for high-performance computing at IDC, a market research firm. The company is wisely “riding the trend” toward low-cost, high-bandwidth clustering using commodity components, he said. Indeed, a team of researchers at Virginia Tech built their 10.2 Tflop/s cluster in a few weeks for $5.2 million, while Japan’s 35.9 Tflop/s Earth Simulator, currently the fastest system in the world, cost around $350 million.
Appearing on the Top 500 list may signal Apple’s official arrival on the supercomputing stage, but the company isn’t counting on a wave of home-grown monster G5 compute farms to build its future cluster business. Rather, the company considers its target market to be smaller, “departmental clusters” built with the Xserve rack-optimized servers it first released in May 2002. Earlier this year, Apple released a new configuration of the Xserve that was “designed for cluster computing, especially in bioinformatics and rendering,” said Alex Grossman, director of hardware storage at Apple. So far, Grossman said, the company is seeing “pretty brisk” sales in the bioinformatics sector for systems in the range of 6 to 50 Xserve cluster nodes with one head node.
“We very rarely see systems above 100 [Xserves], although I wouldn’t say that’s impossible — it’s just not where the sweet spot is,” Grossman said.
Apple doesn’t track its server sales by vertical market, and its most recent earnings report didn’t break out Xserve revenues at all, but Humphreys said that the server is gaining traction in the HPC sector. According to IDC, Apple’s Xserve revenues for the three quarters of 2002 in which the product was available were $35 million. In the first two quarters of 2003, Xserve revenues were already $28 million. Humphreys said this translates into 380 percent year-over-year growth and a run rate of $15 million per quarter.
“We’re seeing a lot of validation out there,” said Grossman. “When we introduced the Xserve a lot of people put their eyebrows up and said, ‘Hmm, this is a low-cost Apple product; how can they do it? And what about Mac OS X vs. Linux?’ And we’ve probably turned that around. If you look at the top 10 supercomputer list … I didn’t see anything that runs Microsoft Windows up there. Everything is running some version of Unix or Linux, so it kind of validates our strategy that you can take this off-the-shelf operating system and run it in one of the top ten supercomputers, and you’re not patching it every day or worrying about the kernel of the week.”
While the company wouldn’t provide sales figures for the bioinformatics sector, “what we’ve seen in bioinformatics has been surprising to us,” said Grossman. “We had estimated a certain amount of our business going there, and we’ve seen a much higher percentage of it, so the whole concept of clusters and Xserve has been better than what we expected.” While a large part of the company’s Xserve sales have been in rendering and the commercial markets, “really where we’re seeing it is in bioinformatics,” Grossman said. “It’s feverish there.”
Building a Community
Apple is building a network of resellers and consulting firms to expand its reach into the bioinformatics market, and also relies on a growing list of commercial and public-sector software packages that support the Mac OS X operating system (see below).
One consulting firm that has raised Apple’s profile in the life science market is BioTeam, which has so far installed around 20 Apple bioinformatics clusters, according to Stan Gloss, who heads up sales and marketing for the company. Gloss agreed with Grossman that Apple is finding a niche with small- to medium-sized clusters. “The average is around 12 nodes,” Gloss said, “and they range from around 2 to 32.”
Grossman credited products like BioTeam’s iNquiry — which automates the process of setting up a biocluster — for the growing popularity of Apple-based systems in bioinformatics. “They ship it on an iPod, and you take the iPod and plug it in through the FireWire port on the iPod and the FireWire port on the front of the Xserve, and it basically goes out and configures your cluster automatically,” Grossman said.
BioTeam is currently in discussions with Apple regarding a reseller agreement for iNquiry.
Although its first sales of Xserve clusters were to academic and government groups, Gloss said the firm is now seeing interest from pharmaceutical and biotech companies, particularly in the smaller systems that Gloss termed “under-the-desk clusters.” Ease of installation has been a barrier for these smaller systems in the past, Gloss said, because the “effort to configure a two-node cluster is the same as it is to configure a 20-node cluster, so it was hard to justify the cost of a smaller system.” Now, he said, the lower costs of the Apple hardware combined with quick and easy installation enable a decent-sized cluster for under $15,000 instead of hundreds of thousands of dollars.
Grossman said that in the academic market, “it’s usually an ‘up-from’ strategy — they’re coming up from some older Intel system into [Xserve].” In pharma, however, “they’re usually coming down from a Sun-based infrastructure or some other big iron, they’re looking to save money, and they want to stay with a Unix-based infrastructure.” In many cases, the transition to an Apple cluster has been spurred by scientists running Unix workstations who have adopted PowerBooks as their portable systems, and then turn to Xserve when they’re in the market for a cluster, Grossman said.
Grossman was unable to discuss whether Apple plans to release a G5 Xserve, but that’s not stopping the higher-powered desktop system from finding its way into high-performance bioinformatics computing environments. Gloss said that BioTeam is currently installing a 16-node cluster of G5 desktops — in the same vein as the Virginia Tech X system — at the Arctic Supercomputer Center at the University of Alaska, Fairbanks. The G5 system is expected to be ready before the end of the year, Gloss said.
— BT
Bioinformatics Applications on Mac OS X
Commercial Packages:
- Accelrys — MacVector
- BioDiscovery — ImaGene
- BioTeam — iNquiry
- Daylight Chemical Informations Systems — Multiple toolkits
- DNAStar — Lasergene
- Electric Genetics — eVoke
- GeneData — Phylosopher, Impressionist
- GenOps — Ngene
- Geospiza — FINCH
- Improvision — Volocity
- Incyte — Bioknowledge Retriever
- InforMax — VectorNTI Suite
- Molecular Biology Insight — Oligo
- NuGenesis — SDMS
- OpenEye — Cheminformatics Suite
- Silicon Genetics — GeneSpring
- Textco — Gene Construction Kit
- The MathWorks — MatLab
- TurboWorx — Builder, Group & Enterprise
Free/Open Source Packages:
- Amber
- Apollo Genome Browser
- Apple-Genentech Blast
- AutoDock Toolkit
- BioPerl, BioJava, BioXML, BioCorba, BioRuby, BioPython
- ClustalW
- Cn3D
- EMBOSS
- Fasta
- Glimmer2
- GROMACS
- HMMer
- MUMmer
- NCBI Toolkit
- OmniGene/OmniView
- PAUP
- Phred-Phrap-Consed
- Phylip
- Primer3
- PyMOL, MolMol, RasMOL, VMD
- Python Molecule Viewer
- Smith-Waterman
- Wise2 (Protein comparison)
- WU-Blast