"DNA sequences from individuals can be prepared so quickly and cheaply now, we can understand evolutionary relationships more accurately than ever before," according to Miller. "The problem is, the number of computations required grows quickly as the amount of data grows. There are only three possible relationships between any four individuals, but there are more than two million different relationships between 10 individuals. A computer that could analyze a million trees per second would require about 20 billion years to test all the possible relationships for just 22 individuals!"
Solving this problem is where the CIPRES Gateway and TeraGrid supercomputers come in. The power of supercomputers comes from parallel computing, in which large analyses are broken into smaller pieces that are run simultaneously on many processor cores. Under the TeraGrid's Advanced User Support program, Wayne Pfeiffer, a distinguished scientist at SDSC, helped improve the parallel performance of RAxML and MrBayes, two widely used phylogenetics codes.
"Most RAxML analyses submitted to the CIPRES Gateway now run on 60 cores of Trestles," said Pfeiffer. "With a typical speedup over a single core of about 30, this means that analyses that would require a month on a laptop can be completed in a day via the gateway."
"This is an excellent example of how science is being transformed through new ways of leveraging the capabilities of today's supercomputers," said Richard Moore, SDSC's deputy director. "Significantly reducing the time it takes researchers to run such complex analyses, while freeing them from having to fully understand all the intricacies of today's supercomputers, means greater scientific productivity. This is what makes the CIPRES Gateway such a valuable phylogenetic resource."
Although SDSC's CIPRES Gateway has been in operation for a little more than a year, it has already provided immed
|Contact: Jan Zverina|
University of California - San Diego