Researchers at IBM and Intel are questioning the methods by which the supercomputing Top500 list is compiled. "The Top500 [uses] an artificial problem—it doesn't measure about 80 percent of the workloads" that are usually run on supercomputers, says Intel's John Hengeveld.
The main problem that some researchers have with the Top500 list is the single metric used to measure the supercomputers, called Linpack. It is hard to determine the actual application speed from the current benchmark, according to an attendee at the Top500 awards ceremony. Linpack is best suited to measure the computational power of a machine, especially in solving linear equations. However, the metric is not very good at estimating the memory performance of a supercomputer, which is an important part of many data-styled problems, Hengeveld says. Another drawback is that Linpack does not measure reliability, says IBM's Dave Turek.
In recent years, many researchers have developed new metrics for measuring supercomputers, such as Virginia Tech's Green500, which measures energy efficiency, and the new Graph500, which measures performance for data-intensive applications.
View Full Article
Abstracts Copyright © 2010 Information Inc., Bethesda, Maryland, USA