超级计算机和化石改变了系统树

[编者的话]

美国国家历史博物馆的研究人员用560个Intel Pentium CPU组合成一个计算机集群，配合化石资料，研究节肢动物的进化，得出了一些非常有意思的结果。这项研究要求强大的计算力量，在他们的CPU集群上要耗费几个月的机时，但是在普通计算机上则要花费46年的时间。

by Rabiya S. Tuma

More data points means more computer time ， and in evolutionary biology the number of possible outcomes for even a relatively small phylogenetic tree is astronomical. To satisfy this unruly demand for computer time ， scientists at the American Museum of Natural History this week unveiled the 9th fastest cluster computer in the world. Now analyzing datasets much larger than previously possible， the scientists there have novel insights into arthropod evolution. Ward Wheeler ， co-director of the Molecular Systematics Laboratories and curator of the Division of Invertebrate Zoology at the Museum ， and his graduate students built the computer by stringing together 560 Intel Pentium CPUs. According to Wheeler， building your own parallel computer using off-the-shelf components ， as they did， is the cheapest way to get a supercomputer.

Speaking at a two-day conference at the Museum called "New Directions in Cluster Supercomputing ，" Wheeler described the computer and his newest efforts to describe the evolution of arthropods ， which includes insects， lobsters， myriapods（centipedes and millipedes ）， and crustaceans （crabs and lobsters）。

Thus far， most scientists in the field of arthropod evolution have assumed that the lineage colonized land once， and that the ancestor that did so gave rise to all of the terrestrial arthropods ， like insects and myriapods. But Wheeler's new phylogenetic trees suggest that the beasts invaded land not once ， but twice， and that insects are more closely related to crustaceans than they are to other terrestrial arthropods.

Wheeler arrived at this new conclusion by including fossil species in his analysis. If he leaves them out，he gets the same conclusion others do - that all terrestrial arthropods share a recent ancestor. He's not sure which answer is correct ， but will likely expand the dataset to include more fossils and see what happens.

But even the existing datasets required several months of continuous computer time on their cluster computer. If they had had to run the experiment on a standard computer， it would have taken 46 years. Thus Wheeler concludes， "We're able to attack investigations that we couldn't take on without high-intensity computing."

The new computer is "very unique，" said Gonzalo Giribet， who was a postdoctoral fellow with Wheeler and is now an assistant professor of evolutionary biology at Harvard University. "It is unique enough that I still use it even though I'm at Harvard." He's also starting to put together a smaller version of the cluster in Boston.

Mark Seager ， assistant department head for terascale systems at the Lawrence Livermore National Laboratories in Livermore ， California ，said that he thinks that computing is not only changing how science is done or how fast it is done ，but also what science is. In his view it used to be composed of experiments and theory ， now it is experiments， theory and computing that make up the field.