Improved Algorithms Speed It Up for Codes

Hazi, A.

Lawrence Livermore National Laboratory

关键词: 75 Condensed Matter Physics, Superconductivity And Superfluidity;    Computers;    Nuts;    Monte Carlo Method;    Physics;   

DOI  :  10.2172/883730
RP-ID  :  UCRL-TR-215720
RP-ID  :  W-7405-ENG-48
RP-ID  :  883730

美国|英语

来源: UNT Digital Library

【 摘 要 】

Huge computers, huge codes, complex problems to solve. The longer it takes to run a code, the more it costs. One way to speed things up and save time and money is through hardware improvements--faster processors, different system designs, bigger computers. But another side of supercomputing can reap savings in time and speed: software improvements to make codes--particularly the mathematical algorithms that form them--run faster and more efficiently. Speed up math? Is that really possible? According to Livermore physicist Eugene Brooks, the answer is a resounding yes. ''Sure, you get great speed-ups by improving hardware,'' says Brooks, the deputy leader for Computational Physics in N Division, which is part of Livermore's Physics and Advanced Technologies (PAT) Directorate. ''But the real bonus comes on the software side, where improvements in software can lead to orders of magnitude improvement in run times.'' Brooks knows whereof he speaks. Working with Laboratory physicist Abraham Szoeke and others, he has been instrumental in devising ways to shrink the running time of what has, historically, been a tough computational nut to crack: radiation transport codes based on the statistical or Monte Carlo method of calculation. And Brooks is not the only one. Others around the Laboratory, including physicists Andrew Williamson, Randolph Hood, and Jeff Grossman, have come up with innovative ways to speed up Monte Carlo calculations using pure mathematics.