05-11-2016, 03:42 PM
1465469658-thesispaper.odt (Size: 44.33 KB / Downloads: 4)
Abstract
Brownmove is a simulation package which is used for the simulation of molecular dynamics and Brownian dynamics (BDS).It has been used to handle many-molecule problems with large number of particle numbers. How ever, this package uses serial computation techniques. Time needed for the simulation of smaller system is larger and time scale and length scale gap between is even larger in all these cases. Therefore a novel parallel approach for the brownmove simulation package is needed. Developing the parallelization of the code is to reduce the processing time, since the time required to analyse the large amount of data is long. It is built using a modular object-oriented programming method for the easy maintenance and extension of the software. In the proposed system we plan to develop a parallel simulation scheme and implement it using Message Passing Interface. Performing a large scale dynamics simulations on a distributed memory parallel
computer is done. A message-passing paradigm is adopted for interprocessor communications using MPI (Message Passing Interface)for the above said parallelization techniques.
Introduction
Traditionally serial computing were used to write the software. In serial computing a problem is divided into discrete chunks or instructions and each instructions are executed sequentially one after another. In serial computing a single problem will be executed at any moment of time. The limitations of the serial programming such as transmission speed, efficiency, use of the underlying parallel hardware had introduced the concept of parallelism. As a result parallelism is ubiquitous nowadays. Parallel computers are variant from conventional serial computer in many ways. In Parallel computing a task is divided into parts and each part is again subdivided into a set of instructions. Each instructions are executed simultaneously by different processor at the same time. Parallel computer architecture consist of the collection of processors which are linked by an interconnection of networks. The advantage of the parallel computing is to save the time and cost, make use of the local resources, make use of the underlying parallel hardware and concurrency etc.
Processing of the big data is very difficult since it is reaching the size of Terabytes and Petabytes.Processing of this large data require weeks or months if one use their own workstations.One of such big data is the brownian dynamics simulation of the many particle system with large number of the particles.Botanist named Robert Brown examined the microscopic zigzag motion of pollen grain in water.Even though he could not explain the reason for this motion.Later,Einstein explained this observations. Einstein found that it is the stochastic process made the polleln grain in water to move. Einstein's idea was then later known as the Brownian Dynamics simulation technique .Few years ago there was no general purpose software for variable particle simulations.Only the existing tools UHBD,SDA,BROWNDYE were there.The goal of these software is to calculate binary encounters,Hydro suite,diffusional properties of the molecule etc.So a new tool known as Brownmove is introduced by DR.Tihamer Geyer.This is a simulation package needed for the Brownian Dynamics and Langevin Dynamics for variable particle which manages different number of rigid and bead spring particles with different type of interactions such as electrostatics,vanderwaals and hydrodynamic interactions.The time required for the simulation using the software will be large since it uses serial compuatioin scheme.
To process this large number of data a parallel techniques and high performance computing is applied for decreasing the total time required for processing.In this paper we aim to introduce a
parallelisation techniques for the brownmove software with message passing Interface.
Message Passing and Parallel Programming
In parallel computing,object oriented programming a form of communication is used known as the message passing. In message passing objects can send and receive messages (comprising zero or bytes, complex data structures, or even segments of code) to other processes.processes also synchronize while waiting for the message.protocol for parallel programming are the one that has become the standard, one which used to be the standard, and one which somehow feel might be the next big thing. Important methodology for the parallel programming are PVM and MPI.Out of these Mpi is the most appropriate one beacause of its efficiency,capability,portability,standardization,functional capacity,performance and availability.This can also be implemented on differ- ent platforms.MPI is standard for writing message passing programmes.Independent process can send messages to each other.
Message Passing Models
Message passing models are similar to the physical attributes of multipro- cessor architecture.Popular implementation of the MPI are MPICH and it is the successor of MPICH2.It extends all the functionalities.The programmes for MPICH are written in C or FORTAN and are linked with the MPI libraries.MIMD is the parallel architecture that is used to run the MPI ap- plications.MPI uses all the processor and the nodes.This MPI programm is well established for writing parallel computing.MPI is a library for all platforms such as linux,windows,os x and act as an interface for many languages such as C,C++,FORTAN etc.MPI can be defined as the ”Message PassingInterface”, and parallelize computational work with help of tools which is used to solve the problem, and then share the result by passing messages. MPI parallelize the programs locally, by running all processes locally or itwill parallelize programs across a cluster, by executing one or more processes per node. Basic MPI Calls There are two views of MPI.First one is that an lightweight protocol with 6 commands. The other one is a depth protocol with hundreds commands. The 6 Basic MPI Commands
1.MPI Init
2.MPI Comm size
3.MPI Comm rank
4.MPI Send
5.MPI Recv
6.MPI Finalize
The detailed description are
1. MPI Init(int *argc, char *argv) Takes the command line arguments
to a program and passes remaining command line arguments to the main
program.
2. MPI Comm size( MPI Comm comm, int *size ) Determines the size of
a MPI Communicator. A communicator is a set of processes that work in
groups. This act as an communicator for all processes available to an MPI
program.
3. MPI Comm rank( MPI Comm comm, int *rank ) Determines the rank
of the current process. The rank can be an integer from 0 to N-1.
4. MPI Send( void *buf, int count, MPI Datatype datatype, int dest, int
tag, MPI Comm comm ) Send the content of the buf, which will count the
elements of type datatype.
5.MPI Recv( void *buf, int count, MPI Datatype datatype, int source,
int tag, MPI Comm comm, MPI Status *status ) Read into buf count values
of type data type
6.MPI Finalize() Terminates the MPI programming
II EXISTING TOOLS
Existing tool for doing Brownian Dynamics simulation are UHBD,MACRODOX,BROWNDYE,BROWNMOVE,SDA etc.One of such software is macrodox which was developed by Scott Northrup and colleagues.Simple atomic detailed BDS model uses this software.This is available at http://iweb.tntech.edu/macrodox/macrodox. html.This software is used to find out the approximate value of PKa.Another software is UHBD and is available at http://adrik.bchs.uh.edu/uhbd.html. This can be used to simulate steady and non steady association rate between the protein protein association.Instead of atomic model they uses simpler model.Another software named SDA was made by Gabdouline and Wade has the ability to compute effective charge and solve simple or atomic model.The software package known as Browndye, is used for simulating the diffusional encounter of two large biological molecules,to estimate second-order rate constants ,encounter probabilities, and explore reaction trajectories.
a.UHBD
This paper is a follow-up to the initial communication (Comput. Phys. Commun. 62 (1991) 187–197) on the Brownian Dynamics/Electrostatics program UHBD developed at the University of Houston. The program is now capable of computing pKas of ionizable groups in proteins, performing Brownian dynamics simulations with a flexible substrate and target, and molecular mechanics/dynamics calculations using a continuum solvent. These new capabilities and other features are discussed along with selected applications which illustrate the capabilities of the current version of UHBD.
b.parallelizing UHBD
This paper tells about the parallelization of the UHBD program which consist of two stages.First is that in this using a macromolecule in solution electrostatic problem is solved and then diffusion of the molecules is stimulated by the BDS.The most difficult section is the parallel computation of the electrostatic forces.In this paper a parallel preconditioner for conjugate gradient is described.Then report the exeperimental result with help of larger number of process.This approach is same as that of domain decomposition method.
b.BROWNDYE
The Browndye software package have two simulation programs and 25 auxiliary programs for the data processing. It computes the 2nd-order rate constant of the association of the two rigid bodies interactions according to Brownian dynamics (BD).It also compute the probabilities of the two bodies association from one binding mode to another,and also compute the molecules’ trajectories. It has features very similar to the packages SDA and MacroDox and like them mainly used for the
simulations of large biological molecules. Its current limitations is that computing the
approximation of flexible molecules and the accurate nature of the force
computations between the molecules. It also draws ideas from UHBD
c.Macrodox
MacroDox is a Brownian Dynamics Software. MacroDox is software for setting up and running Brownian dynamics simulations used for the calculation of the rate of attainment of optimum docking conformations of protein-protein, protein-DNA, and other complexes. This software allows the user to compute and show the physical properties of individual proteins and nucleic acids.Such properties require electrostatic fields, dipole moment, radius of gyration, titration curves, protonation status of titratible sites, and solvent accessible surfaces. This information is used by application softwar to generate a docked complexes using a Brownian dynamics algorithm.This program also compute the bimolecular diffusive rate constant.
d.SDA
Simulation of Diffusional Associatio is used for the protein-protein interactions.This program is divided into three sections.First part is done with the initialization which will load all the data such as potential grid,DX_grid files etc.The second step includes the processing of the data.In this step all the constants were written and the coordinates were shifted to the computed centre after reading from the input center.The third part is to compute the torque,force of selected atom .This step is repeated again.
e.Parallel SDA.
This paper aims at improving the parallelization of the source code to discover the moderm multicore shared memory architecture.This software aims at the study of the biomolecular association.this software is built using object oriented concept to find out the bimolecular protein-protein interactions.This software allows the easy extention and flexible maintenance.This software is also extended to study the features such as transfer rates, predict the structures of biomacromolecular flexible solute.
f.brownmove
Brownmove is a simulation package that is used for managing the many particle systems,This package is built using a modular object oriented concept.The hierarcy of this software is to that it consist of the protein object which can be made from rigid subunits known as the gestalt.Gestalt is also reprensented as the object and this object comntain one or more shape instances.this software consist of the geomshape object which convert all the forces and torque obtained into the displacement.In this paper the source code is implemented sequentially.The disadvantage of this software is that since it uses serial computation method the time required for the simulation is very large.