Barry L Rountree
Barry's current research focuses on using MPI to parallelize heavyweight debugging tools. While it's possible to use binary rewriting to help track down a number of hard-to-diagnose bugs, the overhead required generally means these are used only as a last resort on serial systems. On large-scale parallel systems where machine time is a scarce resource, this approach is currently justifiable for only the most mission-critical codes. However, if MPI ranks can be cloned on several processors it becomes possible to amortize this overhead over the population of clones. For example, parallelizing valgrind's memcheck functionality allows each of n clones to instrument 1/n memory read instructions. Other potential applications are hardware fault and integer overflow detection and performance counter sampling.
Barry additionally work on power-aware supercomputing, using a combination of performance counters, runtime systems and linear programming to route electrical power to where it will have the greatest impact on performance.
Barry holds a Ph.D. in Computer Science from the University of Arizona, an M.S. in Computer and Network Administration from Florida State University, and a B.A. in Theater from the Honors Tutorial College at Ohio University.
Personal Web Site
This personal Web site is not sanctioned or supported by DoE, UC, LLNL, or any other organization.