1 |
|
\appendix |
2 |
< |
\chapter{\label{chapt:oopse}Object-Oriented Parallel Simulation Engine (OOPSE)} |
2 |
> |
\chapter{\label{chapt:oopse}Object-Oriented Parallel Simulation Engine} |
3 |
|
|
4 |
|
Designing object-oriented software is hard, and designing reusable |
5 |
|
object-oriented scientific software is even harder. Absence of |
14 |
|
coordination to enforce design and programming guidelines. Moreover, |
15 |
|
most MD programs also suffer from missing design and implement |
16 |
|
documents which is crucial to the maintenance and extensibility. |
17 |
+ |
Along the way of studying structural and dynamic processes in |
18 |
+ |
condensed phase systems like biological membranes and nanoparticles, |
19 |
+ |
we developed and maintained an Object-Oriented Parallel Simulation |
20 |
+ |
Engine ({\sc OOPSE}). This new molecular dynamics package has some |
21 |
+ |
unique features |
22 |
+ |
\begin{enumerate} |
23 |
+ |
\item {\sc OOPSE} performs Molecular Dynamics (MD) simulations on non-standard |
24 |
+ |
atom types (transition metals, point dipoles, sticky potentials, |
25 |
+ |
Gay-Berne ellipsoids, or other "lumpy"atoms with orientational |
26 |
+ |
degrees of freedom), as well as rigid bodies. |
27 |
+ |
\item {\sc OOPSE} uses a force-based decomposition algorithm using MPI on cheap |
28 |
+ |
Beowulf clusters to obtain very efficient parallelism. |
29 |
+ |
\item {\sc OOPSE} integrates the equations of motion using advanced methods for |
30 |
+ |
orientational dynamics in NVE, NVT, NPT, NPAT, and NP$\gamma$T |
31 |
+ |
ensembles. |
32 |
+ |
\item {\sc OOPSE} can carry out simulations on metallic systems using the |
33 |
+ |
Embedded Atom Method (EAM) as well as the Sutton-Chen potential. |
34 |
+ |
\item {\sc OOPSE} can perform simulations on Gay-Berne liquid crystals. |
35 |
+ |
\item {\sc OOPSE} can simulate systems containing the extremely efficient |
36 |
+ |
extended-Soft Sticky Dipole (SSD/E) model for water. |
37 |
+ |
\end{enumerate} |
38 |
|
|
39 |
|
\section{\label{appendixSection:architecture }Architecture} |
40 |
|
|
41 |
+ |
Mainly written by \texttt{C/C++} and \texttt{Fortran90}, {\sc OOPSE} |
42 |
+ |
uses C++ Standard Template Library (STL) and fortran modules as the |
43 |
+ |
foundation. As an extensive set of the STL and Fortran90 modules, |
44 |
+ |
{\sc Base Classes} provide generic implementations of mathematical |
45 |
+ |
objects (e.g., matrices, vectors, polynomials, random number |
46 |
+ |
generators) and advanced data structures and algorithms(e.g., tuple, |
47 |
+ |
bitset, generic data, string manipulation). The molecular data |
48 |
+ |
structures for the representation of atoms, bonds, bends, torsions, |
49 |
+ |
rigid bodies and molecules \textit{etc} are contained in the {\sc |
50 |
+ |
Kernel} which is implemented with {\sc Base Classes} and are |
51 |
+ |
carefully designed to provide maximum extensibility and flexibility. |
52 |
+ |
The functionality required for applications is provide by the third |
53 |
+ |
layer which contains Input/Output, Molecular Mechanics and Structure |
54 |
+ |
modules. Input/Output module not only implements general methods for |
55 |
+ |
file handling, but also defines a generic force field interface. |
56 |
+ |
Another important component of Input/Output module is the meta-data |
57 |
+ |
file parser, which is rewritten using ANother Tool for Language |
58 |
+ |
Recognition(ANTLR)\cite{Parr1995, Schaps1999} syntax. The Molecular |
59 |
+ |
Mechanics module consists of energy minimization and a wide |
60 |
+ |
varieties of integration methods(see Chap.~\ref{chapt:methodology}). |
61 |
+ |
The structure module contains a flexible and powerful selection |
62 |
+ |
library which syntax is elaborated in |
63 |
+ |
Sec.~\ref{appendixSection:syntax}. The top layer is made of the main |
64 |
+ |
program of the package, \texttt{oopse} and it corresponding parallel |
65 |
+ |
version \texttt{oopse\_MPI}, as well as other useful utilities, such |
66 |
+ |
as \texttt{StatProps} (see Sec.~\ref{appendixSection:StaticProps}), |
67 |
+ |
\texttt{DynamicProps} (see |
68 |
+ |
Sec.~\ref{appendixSection:appendixSection:DynamicProps}), |
69 |
+ |
\texttt{Dump2XYZ} (see |
70 |
+ |
Sec.~\ref{appendixSection:appendixSection:Dump2XYZ}), \texttt{Hydro} |
71 |
+ |
(see Sec.~\ref{appendixSection:appendixSection:hydrodynamics}) |
72 |
+ |
\textit{etc}. |
73 |
+ |
|
74 |
|
\begin{figure} |
75 |
|
\centering |
76 |
|
\includegraphics[width=\linewidth]{architecture.eps} |
77 |
< |
\caption[The architecture of {\sc oopse}-3.0] {The architecture of |
78 |
< |
{\sc oopse}-3.0.} \label{appendixFig:architecture} |
77 |
> |
\caption[The architecture of {\sc OOPSE}] {Overview of the structure |
78 |
> |
of {\sc OOPSE}} \label{appendixFig:architecture} |
79 |
|
\end{figure} |
80 |
|
|
81 |
|
\section{\label{appendixSection:desginPattern}Design Pattern} |
112 |
|
|
113 |
|
As one of the latest advanced techniques emerged from |
114 |
|
object-oriented community, design patterns were applied in some of |
115 |
< |
the modern scientific software applications, such as JMol, OOPSE |
116 |
< |
\cite{Meineke05} and PROTOMOL \cite{Matthey05} \textit{etc}. |
115 |
> |
the modern scientific software applications, such as JMol, {\sc |
116 |
> |
OOPSE}\cite{Meineke05} and PROTOMOL\cite{Matthey05} \textit{etc}. |
117 |
> |
The following sections enumerates some of the patterns used in {\sc |
118 |
> |
OOPSE}. |
119 |
|
|
120 |
|
\subsection{\label{appendixSection:singleton}Singleton} |
121 |
< |
The Singleton pattern ensures that only one instance of a class is |
122 |
< |
created. All objects that use an instance of that class use the same |
123 |
< |
instance. |
121 |
> |
The Singleton pattern not only provides a mechanism to restrict |
122 |
> |
instantiation of a class to one object, but also provides a global |
123 |
> |
point of access to the object. Currently implemented as a global |
124 |
> |
variable, the logging utility which reports error and warning |
125 |
> |
messages to the console in {\sc OOPSE} is a good candidate for |
126 |
> |
applying the Singleton pattern to avoid the global namespace |
127 |
> |
pollution.Although the singleton pattern can be implemented in |
128 |
> |
various ways to account for different aspects of the software |
129 |
> |
designs, such as lifespan control \textit{etc}, we only use the |
130 |
> |
static data approach in {\sc OOPSE}. {\tt IntegratorFactory} class |
131 |
> |
is declared as |
132 |
> |
\begin{lstlisting}[float,caption={[A classic Singleton design pattern implementation(I)] Declaration of {\tt IntegratorFactory} class.},label={appendixScheme:singletonDeclaration}] |
133 |
|
|
134 |
+ |
class IntegratorFactory { |
135 |
+ |
public: |
136 |
+ |
static IntegratorFactory* getInstance(); |
137 |
+ |
protected: |
138 |
+ |
IntegratorFactory(); |
139 |
+ |
private: |
140 |
+ |
static IntegratorFactory* instance_; |
141 |
+ |
}; |
142 |
+ |
\end{lstlisting} |
143 |
+ |
The corresponding implementation is |
144 |
+ |
\begin{lstlisting}[float,caption={[A classic Singleton design pattern implementation(II)] Implementation of {\tt IntegratorFactory} class.},label={appendixScheme:singletonImplementation}] |
145 |
+ |
|
146 |
+ |
IntegratorFactory::instance_ = NULL; |
147 |
+ |
|
148 |
+ |
IntegratorFactory* getInstance() { |
149 |
+ |
if (instance_ == NULL){ |
150 |
+ |
instance_ = new IntegratorFactory; |
151 |
+ |
} |
152 |
+ |
return instance_; |
153 |
+ |
} |
154 |
+ |
\end{lstlisting} |
155 |
+ |
Since constructor is declared as {\tt protected}, a client can not |
156 |
+ |
instantiate {\tt IntegratorFactory} directly. Moreover, since the |
157 |
+ |
member function {\tt getInstance} serves as the only entry of access |
158 |
+ |
to {\tt IntegratorFactory}, this approach fulfills the basic |
159 |
+ |
requirement, a single instance. Another consequence of this approach |
160 |
+ |
is the automatic destruction since static data are destroyed upon |
161 |
+ |
program termination. |
162 |
+ |
|
163 |
|
\subsection{\label{appendixSection:factoryMethod}Factory Method} |
70 |
– |
The Factory Method pattern is a creational pattern which deals with |
71 |
– |
the problem of creating objects without specifying the exact class |
72 |
– |
of object that will be created. Factory Method solves this problem |
73 |
– |
by defining a separate method for creating the objects, which |
74 |
– |
subclasses can then override to specify the derived type of product |
75 |
– |
that will be created. |
164 |
|
|
165 |
+ |
Categoried as a creational pattern, the Factory Method pattern deals |
166 |
+ |
with the problem of creating objects without specifying the exact |
167 |
+ |
class of object that will be created. Factory Method is typically |
168 |
+ |
implemented by delegating the creation operation to the subclasses. |
169 |
|
|
170 |
+ |
Registers a creator with a type identifier. Looks up the type |
171 |
+ |
identifier in the internal map. If it is found, it invokes the |
172 |
+ |
corresponding creator for the type identifier and returns its |
173 |
+ |
result. |
174 |
+ |
\begin{lstlisting}[float,caption={[].},label={appendixScheme:factoryDeclaration}] |
175 |
+ |
class IntegratorCreator; |
176 |
+ |
class IntegratorFactory { |
177 |
+ |
public: |
178 |
+ |
typedef std::map<string, IntegratorCreator*> CreatorMapType; |
179 |
+ |
|
180 |
+ |
bool registerIntegrator(IntegratorCreator* creator); |
181 |
+ |
|
182 |
+ |
Integrator* createIntegrator(const string& id, SimInfo* info); |
183 |
+ |
|
184 |
+ |
private: |
185 |
+ |
CreatorMapType creatorMap_; |
186 |
+ |
}; |
187 |
+ |
\end{lstlisting} |
188 |
+ |
|
189 |
+ |
\begin{lstlisting}[float,caption={[].},label={appendixScheme:factoryDeclarationImplementation}] |
190 |
+ |
bool IntegratorFactory::unregisterIntegrator(const string& id) { |
191 |
+ |
return creatorMap_.erase(id) == 1; |
192 |
+ |
} |
193 |
+ |
|
194 |
+ |
Integrator* |
195 |
+ |
IntegratorFactory::createIntegrator(const string& id, SimInfo* info) { |
196 |
+ |
CreatorMapType::iterator i = creatorMap_.find(id); |
197 |
+ |
if (i != creatorMap_.end()) { |
198 |
+ |
//invoke functor to create object |
199 |
+ |
return (i->second)->create(info); |
200 |
+ |
} else { |
201 |
+ |
return NULL; |
202 |
+ |
} |
203 |
+ |
} |
204 |
+ |
\end{lstlisting} |
205 |
+ |
|
206 |
+ |
\begin{lstlisting}[float,caption={[].},label={appendixScheme:integratorCreator}] |
207 |
+ |
|
208 |
+ |
class IntegratorCreator { |
209 |
+ |
public: |
210 |
+ |
IntegratorCreator(const string& ident) : ident_(ident) {} |
211 |
+ |
|
212 |
+ |
const string& getIdent() const { return ident_; } |
213 |
+ |
|
214 |
+ |
virtual Integrator* create(SimInfo* info) const = 0; |
215 |
+ |
|
216 |
+ |
private: |
217 |
+ |
string ident_; |
218 |
+ |
}; |
219 |
+ |
|
220 |
+ |
template<class ConcreteIntegrator> |
221 |
+ |
class IntegratorBuilder : public IntegratorCreator { |
222 |
+ |
public: |
223 |
+ |
IntegratorBuilder(const string& ident) : IntegratorCreator(ident) {} |
224 |
+ |
virtual Integrator* create(SimInfo* info) const { |
225 |
+ |
return new ConcreteIntegrator(info); |
226 |
+ |
} |
227 |
+ |
}; |
228 |
+ |
\end{lstlisting} |
229 |
+ |
|
230 |
|
\subsection{\label{appendixSection:visitorPattern}Visitor} |
231 |
+ |
|
232 |
|
The purpose of the Visitor Pattern is to encapsulate an operation |
233 |
|
that you want to perform on the elements of a data structure. In |
234 |
|
this way, you can change the operation being performed on a |
235 |
< |
structure without the need of changing the classes of the elements |
236 |
< |
that you are operating on. |
235 |
> |
structure without the need of changing the class heirarchy of the |
236 |
> |
elements that you are operating on. |
237 |
|
|
238 |
+ |
\begin{lstlisting}[float,caption={[].},label={appendixScheme:visitor}] |
239 |
+ |
class BaseVisitor{ |
240 |
+ |
public: |
241 |
+ |
virtual void visit(Atom* atom); |
242 |
+ |
virtual void visit(DirectionalAtom* datom); |
243 |
+ |
virtual void visit(RigidBody* rb); |
244 |
+ |
}; |
245 |
+ |
\end{lstlisting} |
246 |
+ |
\begin{lstlisting}[float,caption={[].},label={appendixScheme:element}] |
247 |
+ |
class StuntDouble { |
248 |
+ |
public: |
249 |
+ |
virtual void accept(BaseVisitor* v) = 0; |
250 |
+ |
}; |
251 |
|
|
252 |
< |
\subsection{\label{appendixSection:templateMethod}Template Method} |
252 |
> |
class Atom: public StuntDouble { |
253 |
> |
public: |
254 |
> |
virtual void accept{BaseVisitor* v*} {v->visit(this);} |
255 |
> |
}; |
256 |
|
|
257 |
+ |
class DirectionalAtom: public Atom { |
258 |
+ |
public: |
259 |
+ |
virtual void accept{BaseVisitor* v*} {v->visit(this);} |
260 |
+ |
}; |
261 |
+ |
|
262 |
+ |
class RigidBody: public StuntDouble { |
263 |
+ |
public: |
264 |
+ |
virtual void accept{BaseVisitor* v*} {v->visit(this);} |
265 |
+ |
}; |
266 |
+ |
|
267 |
+ |
\end{lstlisting} |
268 |
|
\section{\label{appendixSection:concepts}Concepts} |
269 |
|
|
270 |
|
OOPSE manipulates both traditional atoms as well as some objects |
290 |
|
DirectionalAtom}s which behaves as a single unit. |
291 |
|
\end{itemize} |
292 |
|
|
293 |
< |
Every Molecule, Atom and DirectionalAtom in {\sc oopse} have their |
293 |
> |
Every Molecule, Atom and DirectionalAtom in {\sc OOPSE} have their |
294 |
|
own names which are specified in the {\tt .md} file. In contrast, |
295 |
|
RigidBodies are denoted by their membership and index inside a |
296 |
|
particular molecule: [MoleculeName]\_RB\_[index] (the contents |
301 |
|
\section{\label{appendixSection:syntax}Syntax of the Select Command} |
302 |
|
|
303 |
|
The most general form of the select command is: {\tt select {\it |
304 |
< |
expression}} |
304 |
> |
expression}}. This expression represents an arbitrary set of |
305 |
> |
StuntDoubles (Atoms or RigidBodies) in {\sc OOPSE}. Expressions are |
306 |
> |
composed of either name expressions, index expressions, predefined |
307 |
> |
sets, user-defined expressions, comparison operators, within |
308 |
> |
expressions, or logical combinations of the above expression types. |
309 |
> |
Expressions can be combined using parentheses and the Boolean |
310 |
> |
operators. |
311 |
|
|
126 |
– |
This expression represents an arbitrary set of StuntDoubles (Atoms |
127 |
– |
or RigidBodies) in {\sc oopse}. Expressions are composed of either |
128 |
– |
name expressions, index expressions, predefined sets, user-defined |
129 |
– |
expressions, comparison operators, within expressions, or logical |
130 |
– |
combinations of the above expression types. Expressions can be |
131 |
– |
combined using parentheses and the Boolean operators. |
132 |
– |
|
312 |
|
\subsection{\label{appendixSection:logical}Logical expressions} |
313 |
|
|
314 |
|
The logical operators allow complex queries to be constructed out of |
390 |
|
Users can define arbitrary terms to represent groups of |
391 |
|
StuntDoubles, and then use the define terms in select commands. The |
392 |
|
general form for the define command is: {\bf define {\it term |
393 |
< |
expression}} |
394 |
< |
|
216 |
< |
Once defined, the user can specify such terms in boolean expressions |
393 |
> |
expression}}. Once defined, the user can specify such terms in |
394 |
> |
boolean expressions |
395 |
|
|
396 |
|
{\tt define SSDWATER SSD or SSD1 or SSDRF} |
397 |
|
|
437 |
|
some or all of the configurations that are contained within a dump |
438 |
|
file. The most common example of a static property that can be |
439 |
|
computed is the pair distribution function between atoms of type $A$ |
440 |
< |
and other atoms of type $B$, $g_{AB}(r)$. StaticProps can also be |
441 |
< |
used to compute the density distributions of other molecules in a |
442 |
< |
reference frame {\it fixed to the body-fixed reference frame} of a |
443 |
< |
selected atom or rigid body. |
440 |
> |
and other atoms of type $B$, $g_{AB}(r)$. {\tt StaticProps} can |
441 |
> |
also be used to compute the density distributions of other molecules |
442 |
> |
in a reference frame {\it fixed to the body-fixed reference frame} |
443 |
> |
of a selected atom or rigid body. |
444 |
|
|
445 |
|
There are five seperate radial distribution functions availiable in |
446 |
|
OOPSE. Since every radial distrbution function invlove the |
496 |
|
their body-fixed frames.} \label{oopseFig:gofr} |
497 |
|
\end{figure} |
498 |
|
|
499 |
+ |
Due to the fact that the selected StuntDoubles from two selections |
500 |
+ |
may be overlapped, {\tt StaticProps} performs the calculation in |
501 |
+ |
three stages which are illustrated in |
502 |
+ |
Fig.~\ref{oopseFig:staticPropsProcess}. |
503 |
+ |
|
504 |
+ |
\begin{figure} |
505 |
+ |
\centering |
506 |
+ |
\includegraphics[width=\linewidth]{staticPropsProcess.eps} |
507 |
+ |
\caption[A representation of the three-stage correlations in |
508 |
+ |
\texttt{StaticProps}]{This diagram illustrates three-stage |
509 |
+ |
processing used by \texttt{StaticProps}. $S_1$ and $S_2$ are the |
510 |
+ |
numbers of selected stuntdobules from {\tt -{}-sele1} and {\tt |
511 |
+ |
-{}-sele2} respectively, while $C$ is the number of stuntdobules |
512 |
+ |
appearing at both sets. The first stage($S_1-C$ and $S_2$) and |
513 |
+ |
second stages ($S_1$ and $S_2-C$) are completely non-overlapping. On |
514 |
+ |
the contrary, the third stage($C$ and $C$) are completely |
515 |
+ |
overlapping} \label{oopseFig:staticPropsProcess} |
516 |
+ |
\end{figure} |
517 |
+ |
|
518 |
|
The options available for {\tt StaticProps} are as follows: |
519 |
|
\begin{longtable}[c]{|EFG|} |
520 |
|
\caption{StaticProps Command-line Options} |
576 |
|
different vectors). The ability to use two selection scripts to |
577 |
|
select different types of atoms is already present in the code. |
578 |
|
|
579 |
+ |
For large simulations, the trajectory files can sometimes reach |
580 |
+ |
sizes in excess of several gigabytes. In order to effectively |
581 |
+ |
analyze that amount of data. In order to prevent a situation where |
582 |
+ |
the program runs out of memory due to large trajectories, |
583 |
+ |
\texttt{dynamicProps} will estimate the size of free memory at |
584 |
+ |
first, and determine the number of frames in each block, which |
585 |
+ |
allows the operating system to load two blocks of data |
586 |
+ |
simultaneously without swapping. Upon reading two blocks of the |
587 |
+ |
trajectory, \texttt{dynamicProps} will calculate the time |
588 |
+ |
correlation within the first block and the cross correlations |
589 |
+ |
between the two blocks. This second block is then freed and then |
590 |
+ |
incremented and the process repeated until the end of the |
591 |
+ |
trajectory. Once the end is reached, the first block is freed then |
592 |
+ |
incremented, until all frame pairs have been correlated in time. |
593 |
+ |
This process is illustrated in |
594 |
+ |
Fig.~\ref{oopseFig:dynamicPropsProcess}. |
595 |
+ |
|
596 |
+ |
\begin{figure} |
597 |
+ |
\centering |
598 |
+ |
\includegraphics[width=\linewidth]{dynamicPropsProcess.eps} |
599 |
+ |
\caption[A representation of the block correlations in |
600 |
+ |
\texttt{dynamicProps}]{This diagram illustrates block correlations |
601 |
+ |
processing in \texttt{dynamicProps}. The shaded region represents |
602 |
+ |
the self correlation of the block, and the open blocks are read one |
603 |
+ |
at a time and the cross correlations between blocks are calculated.} |
604 |
+ |
\label{oopseFig:dynamicPropsProcess} |
605 |
+ |
\end{figure} |
606 |
+ |
|
607 |
|
The options available for DynamicProps are as follows: |
608 |
|
\begin{longtable}[c]{|EFG|} |
609 |
|
\caption{DynamicProps Command-line Options} |
630 |
|
|
631 |
|
\subsection{\label{appendixSection:Dump2XYZ}Dump2XYZ} |
632 |
|
|
633 |
< |
Dump2XYZ can transform an OOPSE dump file into a xyz file which can |
634 |
< |
be opened by other molecular dynamics viewers such as Jmol and VMD. |
635 |
< |
The options available for Dump2XYZ are as follows: |
633 |
> |
{\tt Dump2XYZ} can transform an OOPSE dump file into a xyz file |
634 |
> |
which can be opened by other molecular dynamics viewers such as Jmol |
635 |
> |
and VMD\cite{Humphrey1996}. The options available for Dump2XYZ are |
636 |
> |
as follows: |
637 |
|
|
638 |
|
|
639 |
|
\begin{longtable}[c]{|EFG|} |
663 |
|
& {\tt -{}-refsele} & In order to rotate the system, {\tt -{}-originsele} and {\tt -{}-refsele} must be given to define the new coordinate set. A StuntDouble which contains a dipole (the direction of the dipole is always (0, 0, 1) in body frame) is specified by {\tt -{}-originsele}. The new x-z plane is defined by the direction of the dipole and the StuntDouble is specified by {\tt -{}-refsele}. |
664 |
|
\end{longtable} |
665 |
|
|
666 |
< |
\subsection{\label{appendixSection:hydrodynamics}Hydrodynamics} |
666 |
> |
\subsection{\label{appendixSection:hydrodynamics}Hydro} |
667 |
|
|
668 |
+ |
{\tt Hydro} can calculate resistance and diffusion tensors at the |
669 |
+ |
center of resistance. Both tensors at the center of diffusion can |
670 |
+ |
also be reported from the program, as well as the coordinates for |
671 |
+ |
the beads which are used to approximate the arbitrary shapes. The |
672 |
+ |
options available for Hydro are as follows: |
673 |
|
\begin{longtable}[c]{|EFG|} |
674 |
|
\caption{Hydrodynamics Command-line Options} |
675 |
|
\\ \hline |
682 |
|
-i & {\tt -{}-input} & input dump file \\ |
683 |
|
-o & {\tt -{}-output} & output file prefix (default=`hydro') \\ |
684 |
|
-b & {\tt -{}-beads} & generate the beads only, hydrodynamics calculation will not be performed (default=off)\\ |
685 |
< |
& {\tt -{}-model} & hydrodynamics model (support ``AnalyticalModel'', ``RoughShell'' and ``BeadModel'') \\ |
685 |
> |
& {\tt -{}-model} & hydrodynamics model (supports ``AnalyticalModel'', ``RoughShell'' and ``BeadModel'') \\ |
686 |
|
\end{longtable} |