Data-driven time parallelization

Lei Ji, Yanan Yu, Namas Chandra, Hugh Nymeyer, Ashok Srinivasan

Research output: Chapter in Book/Report/Conference proceedingConference contribution


We present a new approach to parallelization of important scientific applications. It is based on the observation that results of prior, related, simulations are often available. We use such data to parallelize the time domain. We demonstrate the effectiveness of our approach in Molecular Dynamics (MD) simulations, which are widely used in nano and nano-bio sciences. An important limitation of MD is that the time-step size is around a femto-second. So a large number of time-steps are required to simulate to realistic time scales. Conventional parallelization is of limited effectiveness here - the most scalable codes currently are not efficient at granularities finer than several milliseconds per iteration. Using our approach, Carbon Nanotube simulations scale to granularities as fine as around ten microseconds per iteration. We also present results on protein unfolding simulations of AFM pulling, where we obtain additional one order of magnitude scalability over conventional parallelization.

Original languageEnglish (US)
Title of host publicationProceedings of the 2006 ACM/IEEE Conference on Supercomputing, SC'06
StatePublished - 2006
Externally publishedYes

Publication series

NameProceedings of the 2006 ACM/IEEE Conference on Supercomputing, SC'06

All Science Journal Classification (ASJC) codes

  • General Computer Science


Dive into the research topics of 'Data-driven time parallelization'. Together they form a unique fingerprint.

Cite this