Performance models on QCDOC for molecular dynamics with Coulomb potentials

Yuefan Deng, James Glimm, James W. Davenport, Xiaodan Cai, Eunice E. Santos

Research output: Contribution to journalArticlepeer-review

Abstract

We estimate that a novel architecture massively parallel computer, the QCDOC, can integrate molecular dynamics equations for 105 particles interacting via long-range forces (including Coulomb) for 1-10 μs of simulated time using several weeks of computing time using 8000 or 10,000 processors. This number of atoms is typical for biological molecules. The two main conclusions we reach are as follows. (1) This is an increase of more than one order of magnitude in simulated time over current simulations. (2) The novel architecture, with 24 parallel channels of low latency communication per processor, allows improved long-range communication and an unusual degree of fine-scale parallelism, compared to conventional switch-based architectures. The technical heart of the paper is a detailed analysis of the computing time used in the Ewald method as a function of the required accuracy, the size of the molecular dynamics cell, and the hardware design parameters.

Original languageEnglish (US)
Pages (from-to)183-195
Number of pages13
JournalInternational Journal of High Performance Computing Applications
Volume18
Issue number2 SPEC. ISS.
DOIs
StatePublished - Jun 2004
Externally publishedYes

Keywords

  • Ewald summation
  • Molecular dynamics
  • Parallel computing
  • Timing estimates

ASJC Scopus subject areas

  • Software
  • Theoretical Computer Science
  • Hardware and Architecture

Fingerprint

Dive into the research topics of 'Performance models on QCDOC for molecular dynamics with Coulomb potentials'. Together they form a unique fingerprint.

Cite this