J. Comput. Chem. Jpn., Vol. **, No. **, pp. ** ** (****) 2013 Society of Computer Chemistry, Japan - IPDP -, *,,, **, 240-8501 79-5 *e-mail: tamaki-teppei-hb@ynu.jp, **k-ueda@ynu.ac.jp (Received: February 5, 2013; Accepted for publication: May 27, 2013; Advance publication: July 25, 2013) MPI OpenMP ( ) IPDPCharmm キーワード: 1 MPI [1] OpenMP [2] API (Application Program Interface) 2 3 ( ) DOI: 10.2477/jccj.2013-0004 1
CPU MPI OpenMP IPDP IPDP PHP [3] Web 2 2.1 () (Brute Force Attack) CPU 3 (φ ψ) MPI OpenMP ( ) MPI OpenMP 2 J. Comput. Chem. Jpn.
MPI Molecular Dynamics (MD) 2.2 C C#define ( ) IPDP ## 3 IPDP IPDP 2.3 (Node) 1 IPDP 2.4 IPDP PostgreSQL [4] PHP5 Zend Framework [5] libpq HTTP POST Web PHP 1 2.5 IPDP 2 ( ) ( ) IPDP DOI: 10.2477/jccj.2013-0004 3
Table 1. IPDP Operation environments OS Linux (Fedora 13) Web Server Apache 2.2.17 PHP 5.3.6 DB Server (Master only) PostgreSQL 9.0.2 Other software XZ (liblzma) 4.999.9beta Calculation Software 1 1 CPU ( ) CPU () 3 3.1 Web Web PHP IPDP Linux Web (Apache + PHP + PostgreSQL) PHP Zend Framework HTML Smarty [6] gnuplot [7] xz [8] Table 1 gnuplot 4.2p6 Charmm 34 (Slave only) Web Browser (User only) Mozilla Firefox 3 Microsoft Internet Explorer 8 Opera Google Chrome 2 1 1 2 2 PHP Web (PostgreSQL) PHP Web (PostgreSQL) Charmm 3.2 run input_file divided_run output_ file 4 run input_file ## divided_run divided_run divided_ run divided_run output_file divided_run output_file divided_run 3.3 Figure 1 a) Server (Divider) c) Analysis b) Client (Executer) a b 4 J. Comput. Chem. Jpn.
Figure 1. Data flow in IPDP. 4 Charmm 4.1 Charmm Charmm [9] Charmm Charmm goto 4.2 Web IPDP Figure 2 " " " " " " "" " " ID Charmm open (PDB ) IPDP## Charmm set ##phi 0 to 360 step 10 (phi Charmm ) IPDP ## set phi 0 set phi 10 36 (Figure 3) 3 10 1000 PHP DOI: 10.2477/jccj.2013-0004 5
Figure 2. A screenshot of the top page of the project list. Figure 3. An Example of macro expansion for charmm input file on IPDP. Web " " 4.3 IPDPGlucose2 β(1 4) Cellobiose (Figure 4)φ 6 J. Comput. Chem. Jpn.
Figure 4. Structure of cellobiose. Arrows are the torsion angles considered to generate conformers. Figure 5. Conformation energy map of cellobiose calculated by using the analysis module in IPDP. The results are represented in two ways; left is surface type, right is contour type. Contours are in 2 kcal/mol intervals. ψ [10] φ ψ 2 10 φ ψ 0 3601036 6 gauche+, gauche-, trans 1203 36^2*3^10 = 76527504 7650 Charmm goto label if φ ψ 36*36 = 1296 φ ψ gnuplot (Figure 5) Intel Core i7 (4 core/8 thread) 18 (parallel efficiency) (speed up effect) Figure 6 1 8 6 48 5 IPDP Web PostgreSQL DOI: 10.2477/jccj.2013-0004 7
Figure 6. Parallel efficiency and speed up ratio as a function of the number of nodes. Reference [1] OpenMP http://openmp.org/ [2] MPI (Message Passing Interface) http://www.mpi-forum.org/ [3] PHP (Hypertext Preprocessor) http://www.php.net/ [4] PostgreSQL http://www.postgresql.org/ [5] Zend Framework http://framework.zend.com/ [6] Smarty http://www.smarty.net/ [7] gnuplot http://www.gnuplot.info/ [8] XZ Utils http://tukaani.org/xz/ [9] Charmm (Chemistry at HARvard Macromolecular Mechanics) http://www.charmm.org/ [10] K. J. Naidoo, J. W. Brady, J. Am. Chem. Soc., 121, 2244 (1999). [CrossRef] 8 J. Comput. Chem. Jpn.
Distributed Processing Method on the Calculation for the Search of All Solution Development of the Distributed Processing System IPDP Makoto YABE, Teppei TAMAKI*, Minoru TAKEDA, Junichi KOIZUMI, Kazuyoshi UEDA** Graduate School of Engineering, Yokohama National University, 79-5 Tokiwadai, Hodogaya-Ku, Yokohama 240-8501, Japan *e-mail: tamaki-teppei-hb@ynu.jp; **k-ueda@ynu.ac.jp With the rapid expansion of multi-core computing, the parallel processing platform such as MPI and OpenMP becomes popular as a way to improve the efficiency of computation. These techniques can increase the speed of calculation of software after compiled source codes, with revision using parallel programming rules, through parallel libraries. However, parallel programming is generally difficult for effective speeding-up in the search of finding all solution, because parallel processing is usually repeated each time in the calculation of the energy and structure optimization. To solve these problems, we developed a system based on the distributed processing method, which consists of the procedures of the pretreatment of the division of input-file and distribution of divided processes to many slave machines. This system can divide the input-file into appropriate size of jobs and execute them without having to optimize the source code with parallel processing. It is expected to improve computing speed, reliability of result and data accessibility. To show the efficiency of this system, we applied it to make a conformational energy map of cellulose triacetate using the charmm molecular dynamics simulation program. Keyword: Distributed processing, Search all solution, Docking simulation, Energy map, Database, Preprocessor, Process management DOI: 10.2477/jccj.2013-0004 9