filename    : 385.abstract
entry       : techreport
author      : B. Uk, M. Taufer, T. Stricker, G. Settanni, A. Cavalli
title       : Implementation and Characterization of Protein Folding
              on a Desktop Computational Grid - Is CHARMM a suitable 
              candidate for the United Devices MetaProcessor?
institution : ETH Zürich
number      : 385
year        : 2002
type        : Technical Report
address     : Institute for Comutersystems
month       : October
language    : English
keywords    : Protein folding, widely distributed computing, 
              desktop grids, workload characterization, 
              search algorithms, CHARMM, United Devices MetaProcessor.
pages       : 17
abstract : 

  CHARMM is a popular molecular dynamics code for computational biology.
  For many CHARMM applications such as protein folding, desktop grids
  could become viable alternatives to clusters of PCs.
  
  In this report, we present a prototype and discuss the viability of
  a protein folding application with CHARMM on the United Devices
  MetaProcessor, a platform for widely distributed computing. We
  identify the algorithmic approach of protein folding as a hybrid
  search algorithm with best-first, depth-first and breadth-first
  components and address the issues of task scheduling and fault
  tolerance.
  
  The performance evaluation of our system indicates that the
  calculation is robust against the heterogeneity of compute nodes and
  limited communication capabilities typically found in desktop
  grids. We show that there is an interesting tradeoff between
  accuracy and task parallelism resulting in optimal work-pool size
  for a given platform and a given simulation.  Surprisingly the
  platform heterogeneity of a desktop grid positively affects the
  quality of protein folding simulations.
  
  Protein folding calculations with CHARMM turn out to be well
  suitable for desktop grids like e.g. the United Devices
  MetaProcessor. Our software system can make a large amount of nearly
  free compute cycles available to computational biologists.