How many branches does your high-performance-computing decision tree have?
You’re the IT manager for a genomics startup. Your science team is relying on you to build and maintain a computing infrastructure that meets all its needs: storing, mining, visualizing, integrating, and sharing data; running dozens of bioinformatics software applications as well as BLAST and other compute-intensive algorithmic queries; assembling genomes and predicting protein structures; managing laboratory information; accessing the Internet and using e-mail; and so on. Your system needs to be secure and, of course, it can’t break the bank.
Where do you start? For this special supplement on high-performance computing, Genome Technology lured seven industry experts to a meeting room in San Francisco and asked them to spend an hour discussing just that. Would they call in a consultant? Go directly to one of the big computer vendors? Make a trip to the PC store and rack up a homemade cluster? What about planning for the future? How do you build a system that is flexible enough to accommodate change? As genome data sources expand and computing tasks get more complicated, will your system adapt?
As their discussion and the three case studies that follow indicate, much of computing decision-making considers factors that seem to have little to do with actual computational chores: How big is your staff? What’s your budget? How much floor space do you have? What can your HVAC system handle? How are your users situated geographically? Can just one system serve your needs, or do you need two? Is it really a high-performance system you need? Or, as Blackstone’s Matthew Trunnel suggests, could a high-throughput system satisfy your demands?
Our panelists had the luxury of a thought experiment. Stanley Burt, Jeffrey Skolnick, and Simon Weston have had to face all these questions in real life. Each of their organizations had somewhat different genomic computing demands, and, as you’ll see, each implemented a radically different solution. What choices would you make?
The Editors, Genome Technology