Download presentation
Presentation is loading. Please wait.
Published byHerbert Harvey Modified over 6 years ago
1
Antonio Laganà University of Perugia (UNIPG), IT
HIGH PERFORMANCE GRID COMPUTING FOR MOLECULAR AND MATERIALS SCIENCE AND TECHNOLOGY (CMMST) Antonio Laganà University of Perugia (UNIPG), IT E.Rossi, S. Evangelisti, A. Monari, S. Rampino, M. Verdicchio, C. Angeli, K. Baldridge, G.L. Bendazzoli, S. Borini, R. Cimiraglia, P. Kallay, H.P. Lüthi, K. Ruud, J. Sanchez-Marin, A. Scemama, P. Szalay, A. Tajti, M. Cecchi, M. Carpene, A. Costantini, C. Manuali,E. Garcia, S. Farantos 1
2
SUMMARY 1) DATA AND FLUX TRANSFER IN CMMST APPLICATIONS
2) HTC-HPC FUNCTIONALITIES TRANSFER (IGI-CINECA experimentation) 3) KNOWLEDGE TRANSFER IN CMMST APPLICATIONS 2
3
1) A TYPICAL DATA AND FLUX TRANSFER: THE VIRTUAL EXPERIMENT
3
4
THE CROSSED BEAM (or spectroscopy, NMR, crystallography, etc
THE CROSSED BEAM (or spectroscopy, NMR, crystallography, etc.) VIRTUAL EXPERIMENT - O. Gervasi, A. Lagana’, SIMBEX: a portal for the a priori simulation of crossed beam experiments, Future generation Computer Systems, 20(5), (2004). 4
5
THE GRID EMPOWERED WORKFLOW
NO NO Is there a suitable PES? NO Are ab initio calculations available? Are ab initio calculations feasible? INTERACTION YES YES YES SUPSIM Are dynamics calculations direct? NO Take Force fields from databases FITTING Import the PES routine YES YES DYNAMICS Exact quantum calculations? Approximate quantum calculations? NO NO Semiclassical calculations? NO YES YES YES QDYN TD / TI APPROXIMATE QDYN SEMICLASSICAL SC_IVR QUASI- CLASSICAL OBSERVABLES Fixed J and Energy Fixed Energy Fixed Temperature NO NO NO YES YES YES Scalar and vector correlations Cross-sections Thermal Rate coefficients Thermodynamic properties
6
DATA FORMATS System input Q5Cost Interaction D5Cost Dynamics
Statistics Virtual Monitors 6
7
AB INITIO PACKAGES (Wikipedia)
ABINIT GPL Fortran PW ACES II Acad Fortran GTO ACES II MAB Acad. Fortran GTO ACES III GPL Fortran/C++ GTO ADF Comm Fortran STO ATK Comm C++/Pyton NAO/EHT
8
AB INITIO PACKAGES CADPAC Acad Fortran GTO COLUMBUS Acad Fortran GTO
CRYSTAL Acad Fortran GTO DALTON Acad Fortran GTO DFT GPL C PW/Wavelet DIRAC Acad F77, 90, C GTO GAMESS Acad/Comm Fortran GTO GAUSSIAN Comm Fortran GTO MOLCAS Comm Fortran GTO MOLPRO Comm Fortran GTO NWChem ECLv F77/C GTO, PW TERACHEM Comm C/CUDA GTO
9
DATA FORMATS °QC: FIXED GEOMETRY MULTI STATE - SD (Small data) input, parameters, properties -> formatted human readable like XML - LD (Large data) binary working quantities like 4 index integrals IJKL OPEN BABEL: open, collaborative project allowing search, conversion, analysis, or data storing in CMMST CML: A XML-based format for chemistry HDF5: a hierarchical data model °QD: ALL GEOMETRIES SINGLE STATE (formats as for QC)
10
HDF5 ° CONTAINERS (groups) domains of quantities defined by common basis functions (AO atomic orbitals, MO molecular orbitals, Wave functions WF, ..) ° METADATA (attributes) simple and small sets of data describing a set of data ° DATA (data ses) large set of binary quantities referring to wavefunction integrals storing in matrices with an arbitrary number of indices ° LIBRARIES (Utilities and tools) Q5dump reports the content of Q5cost files Q5edit python based program text editor
11
Determinants/ coefficients
System overlap oneint twoint Title Electrons /a,b AO Name Num_orb_sym Num_orb_tot Labels Transformation MO AO_pointer Num_orb_tot Labels Orbitals SCF_energy Classification Occ_num Symmetry Ctime (s) Atime (s) Q5version (s) Symmetry: - num_sym - labels densities Geometry: - charges - coordinates - nuclear_energy - num_atom - atomic_number Basis Coord system Atom Angular number Magnetic number Coeff/exp Num of contracted Num of primitives prop Description Rank Real/Complex Index/value WF MO_pointer Energy Core Energy Num_dets DetCoeff Determinants/ coefficients Energy derivatives Non adiabatic coupling EXTENSION OF Q5 TO SUPPORT D5
12
INSERTION OF Q5/D5 IN GEMS
Fitting ForceField SMatrix QuantumFunction Trajectories Spectra CrossSection RateCoeffs System Interaction Dynamics Observables AbInitio q5 “file system” + energy derivatives + non-ad. coupling els. – system general info
13
QUANTUM DYNAMICS PACKAGES
ABC RWAVEPR FLUSS MCTDH
14
CLASSICAL DYNAMICS PACKAGES (LARGE SYSTEMS)
VENUS AMBER CHARMM DL_POLY GROMACS MOLDY TINKER YASARA
15
Force field representation
Many body expansion truncated to two-body terms Two body terms are of the atom(ion) – atom(ion) type Portability among different systems
16
2) HTC – HPC FUNCTIONALITIES TRANSFER BETWEEN IGI AND PRACE
17
THE DECOUPLED TRAJECTORY DISTRIBUTION
Define quantities of general use TRAJ Iterate over initial conditions the integration of individual trajectories (ABCTRAJ, etc.) Collect individual trajectory results return 17
18
Define quantities of general
THE DECOUPLED WAVEPACKET DISTRIBUTION Define quantities of general use Iterate over initial conditions the time propagation (RWAVEPR, CYLHYP, etc.) TD Collect single initial state S matrix element return 18
19
DECOUPLING QUANTUM TIME INDEPENDENT
Define quantities of general use including the integration bed TI Iterate over the reaction coor- dinate to build the interaction matrix Collect coupling matrix elements Broadcast coupling matrix Iterate over total energy value the integration of scattering equations return Collect state to state S matrix elements 19
20
THE DISTRIBUTION OF SUPSIM (Ab initio)
Iterate over the system Geometries the call of ab initio suites of codes (GAMESS, GAUSSIAN, MOLPRO, etc) Define the characteristics of the ab initio calculation, the coordinates used and the Variable’s intervals SUPSIM L. Storchi, F. Tarantelli, A. Lagana’, Computing Molecular energy surfaces on the grid, Lecture Notes in Computer Science 3980, (2006). return Collect single molecular geometry energy 20
21
SUPSIM: HTC + HPC (squares) CALCULATIONS
22
MCTDH: HPC (square) + HTC CALCULATIONS
23
GriF: a collaborative tool for grid empowering computational applications
Makes grid applications black box like and the distribution of tasks over the grid automatic. Adopt a collaborative JAVA Service Oriented Architecture (SOA) framework articulated as a set of grid services Provide users with standard operational modalities based on friendly user driven services allowing the composition of one or more services transparently to their implementation details. Conveniently OPTIMIZES THE ROUTING of the jobs to the MOST appropriate HTC OR HPC machine C. Manuali – A. Laganà University of Perugia (IT)
24
Service Oriented organization of GriF: two JAVA servers, one JAVA client
TOP 1,2 service discovery 3-7 program execution on the UI YR= yet a registry YP= yet a provider YC = yet a client BOTTOM grid proxy management and its YC interaction
25
THE GRIF SERVICE ORIENTED APPROACH
A repository (Softscience?) of services A metric to evaluate QoS and QoU A credit award and service cost system A parameterized resources selection C. Manuali, A. Lagana’ GRIF: A New Collaborative Framework for a Web Service Approach to Grid Empowered Calculations Future Generation of Computer Systems, 27(3), (2011) DOI /j.future
26
3) INCORPORATE GEMS INTO KNOWLEDGE OBJECTS (KO)
27
THE DISTRIBUTED REPOSITORY OF KNOWLEDGE OBJECTS
Incorporate GEMS into KOs Reusable, downloadable units of knowledge. KOs stored in distributed repositories. Online indexes of KO with access control. Dynamical evolution of KOs and distributed repositories Linking repositories belonging to different organizations on the Grid to share contents and mutually improve them. So, here’s what we thought: -we could package content into Learning Objects, which are self-contained units of knowledge regarding a particular subject -then we also could store these learning objects into repositories, which are websites that allow authenticated users to browse and download the learning content - finally, if we could make our repositories talk to each other, we could have an environment where learning materials are able to freely flow and be downloaded and used easily 27
28
BASIC FEATURES Core Drupal Modules Freely available community modules
Two (opensource) custom modules Some federates (to begin testing) A database And here’s our final recipe: We took the core drupal modules A couple of freely available community modules We wrote two addittional modules And added some federates and a database to the mix 28
29
KO SW Our results Federation of autonomous repositories
Automatic content sharing Downloadable content (for registered users) Simplified content import from Moodle KO Dependency management SW And here it is what we got from it: a distributed federation with no central server (meaning that all the federates have the same importance) capable of automatic real-time content sharing, allowing users of any federate to browse and download content uploaded to any other federate. To ease the initial population phase, content can be added using an optional Moodle import module, which allows an administrator to transfer existing content from a Moodle installation. Plus, we added dependency management: If you think about Learning Objects from a wider perspective, they could be not only simple text files, slides or pdfs, but they could be more complex objects like 3d virtual worlds and uncommon file formats which will need special software tools to be visualized by the end user. Dependencies allows to create a relationship between the learning object and what we call software attachment in order to tie the two together. In this way the system will ensure that the user always downloads the required software attachments when he or she requests learning objects from the repository. KO 29
30
An implemented use case
KO Chemistry (Perugia) Computer Science (Perugia) The use case I’m going to talk about involves two different repositories: The first one is the computer science repository, where an user with elevated privileges will upload FreeWRL , an opensource X3D viewer, as a software attachment. Then we will add a second repository, belonging to the chemistry department, and we will make both join the same federation. After this step we will see that a user of the chemistry repository willing to download an X3D Learning Object, will also automatically get the needed software attachment. SW User icons by pixel-mixer.com 30
31
NEXT STEPS Stabilize the other sites (Thessaloniki, Madrid, Lyon) and add new ones Improve software Speed up synchronization/recovery, Upgrade online visualizers for particular file formats (pictures, TEX files…) Extend KOs’ lifetime Allow third parties to improve existing KOs Track KOs after the download Allow uploading KO outputs to the Repository Our future plans include to speedup network transmissions and further improve repository reliability in case of network problems We’d also like to intelligently visualize particular content type like tex files We’re currently working on an accessory module to allow others to attach content to existing learning objects We’re also aiming at tracking LO usage after the download with auxiliary software capable of sending back certain informations to the federation on user request We’re also currently working on a way to automatically import content from moodle, wich is a well known lms, to drupal. 31
32
The new use case Chemistry (Perugia) Computer Science (Perugia) KO KO SW It’s not mandatory that the requesting student belongs to the chemistry or the computer science repository, in fact it could be registered to a different repository, as long as it belongs to the same federation. The download procedure is slightly different though. The user will find the learning object listed on the server he belongs to, but the learning object, as we just saw, is not stored there. The second chemistry server will then ask the one in Perugia to authorize the LO download, and in response to this request an authorization token will be issued: this token will be sent back to the requesting server that will redirect the user to the download page on the chemistry repository in perugia, where he will finally download the requested LO along with the needed software attachment. Chemistry (Tessaloniki) SW User icons by pixel-mixer.com 32
33
AIR POLLUTION SIMULATION
CPM10 Concentration from CHIMERE-aerosols 33
34
SUMMARY Flux and (standardized) data transfer
Functionalities integration including HTC-HPC Knowledge transfer, evolution and communication
35
FUNDING AND COLLABORATIONS
CDK group, Dept. Chemistry, Perugia (Crocchianti, Faginas Lago, Pacifici, Skouteris, Costantini, Rampino, Manuali, Verdicchio, Filomia) HPC group, Dept. Math&Inf, Perugia (Gervasi, Tasso) DeciQ and Qdyn groups of COST D37 COMPCHEM Virtual Organization EGI-Inspire, ESA, Computational Chemistry Division of the European Chemical Society TANKS FOR YOUR ATTENTION
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.