The distributed diagonal force decomposition method for parallelizing molecular dynamics simulations

Abstract

Parallelization is an effective way to reduce the computational time needed for molecular dynamics simulations. We describe a new parallelization method, the distributed-diagonal force decomposition method, with which we extend and improve the existing force decomposition methods. Our new method requires less data communication during molecular dynamics simulations than replicated data and current force decomposition methods, increasing the parallel efficiency. It also dynamically load-balances the processors' computational load throughout the simulation. The method is readily implemented in existing molecular dynamics codes and it has been incorporated into the CHARMM program, allowing its immediate use in conjunction with the many molecular dynamics simulation techniques that are already present in the program. We also present the design of the Force Decomposition Machine, a cluster of personal computers and networks that is tailored to running molecular dynamics simulations using the distributed diagonal force decomposition method. The design is expandable and provides various degrees of fault resilience. This approach is easily adaptable to computers with Graphics Processing Units because it is independent of the processor type being used.

Implementation of the force decomposition machine for molecular dynamics simulations

“…The DDFD method [21], which is implemented in the CHARMM program [2], exhibits a characteristic data exchange pattern where nodes communicate with a limited set of other nodes. A set of nodes exchanges set-local data independently from the other sets.…”

“…In the DDFD method, the N atoms of the simulated system are divided into B blocks (disjoint sets). The atoms are randomly assigned to blocks to provide an acceptable inital load balance, though dynamic load balancing also occurs during the simulation [21]. There are B 2 block products formed by pairs of blocks.…”

“…The FDM is designed for the communication patterns of the distributed diagonal force decomposition (DDFD) method [21], a new parallelization method for MD simulation. The goals of the FDM and the implemented architecture are presented in the Network Architecture subsection.…”

Parallel‐ProBiS: Fast parallel algorithm for local structural comparison of protein structures and binding sites

“…[16] The central processing units (CPUs) of modern computers have multiple cores that are separately programmable and can, when used proficiently, offer significant increases in computation speed over single CPUs, [17] and Ethernet connections enable fast connectivity between separate computers. Parallelization has been applied effectively to related problems in computational chemistry, including molecular dynamics [18] and three-dimensional similarity comparisons of small molecules. [19] In this article, we present a parallelized version of ProBiS.…”

“…This is a new algorithm that calculates the local similarity metric between protein structures and is especially suited to efficient execution on multiple CPUs and on computers of varying power interconnected in a network, which are available in contemporary computing platforms or computing clouds. [18] When calculating local structural similarities in a database of over 29,000 protein structures, this parallel version of the ProBiS algorithm is 180 times faster on a cluster of 49 computer nodes than the existing ProBiS algorithm implemented with script-based concurrent runs. We explain the ProBiS algorithm, provide a description of our new Parallel-ProBiS algorithm, and present performance benchmarks and comparison of the two algorithms.…”