Marine Hydrophysical Institute of the National Academy of Science of Ukraine
Marine Hydrophysical Institute (MHI,
www.mhi.iuf.net/eng/) of the National Academy of Science of Ukraine is the member of Ukrainian National Grid (UNG,
infrastructure.kiev.ua/en/about/statute/), and the Institute grid-site (
hpc-mhi.org/?lang=en) is a component of the Ukrainian Grid infrastructure. UNG is the collaboration of the different institutions, which implements grid-technologies into various areas of science, education, medicine, industry and social life. The basic tools of the UNG are the ARC and gLite middleware computer software. In the MHI the work on grid-technology implementation to numerical simulation of dynamical processes in the Black Sea and the Sea of Azov are carried out now. The objective of this work is the development of technique and grid software for the computing of wind wave and sea current parameters in the UNG environment.
The MPI version of SWAN (Simulating Waves NearShore,
www.swan.tudelft.nl/) model is chosen as the basic wind wave model. This allows using the preferences of parallel and grid computing together. To calculate sea current parameters the modified barotropic version of well known POM (Princeton Ocean Model,
www.aos.princeton.edu/WWWPUBLIC/htdocs.pom/) model is used. Due to the difficulties with adaptation of the model to the architecture of each grid-site it is used in the single processor mode.
Problems of Sea Dynamics which Can be Solved in Grid Environment
As the most resource consuming problems of sea dynamics for which the using of grid should give a significant increasing of computing efficiency one may consider for example multiversion calculation of wind waves and currents, their retrospective analysis and ensemble prediction. To effectively solve in grid environment these (and any other) problems it is necessary to make decomposition which means the transformation of the initial problem into sequence of independent sub-problems. In general case the number of sub-problems n is determined by peculiarities of solving problem and productivity of available computing resources. We consider in details the approaches to the decomposition of mentioned above problems which are based on physical peculiarities of wind waves and barotropic currents.
Multiversion calculation of hydrodynamic fields is necessary in cases when a lot of problems of the same type are needed to be solved with different input data. The simplest example is the calculation of stationary wave and current fields in the certain water area for large quantity of velocities and directions of constant in time and homogeneous in space wind. This kind of problems occurs, for instance, in the process of development of data bases of wind wave and current simulation results for different synoptic situations. These data bases are useful for the hydrometeorological support of designing of hydraulic works in the near shore zones. For the each value of wind velocity and direction the problems are completely independent and they may be solved in any order. So the multiversion calculation is the problem which has the parallelism by data and may be effectively solved in grid.
Retrospective analysis (re-analysis) is needed to obtain statistical characteristics and climatic trends of wave and current regimes. Re-analysis data are the result of numerical simulation of hydrodynamical fields for a long period of time (several tens of years) in the certain water area by using of atmospheric fields obtained from corresponding model. It’s clear that for large period of time solving such kind of problem demands significant computing resources. To solve the re-analysis problem in the grid environment the period of computation is divided on n intervals with variable widths. On the each interval wave and current fields are calculated by using corresponding atmospheric conditions. Calculations within the bounds of each interval are started from arbitrary initial conditions (for example, with zero ones). To exclude the effect of initial conditions the intervals of calculations are chosen with some covering, so the end of previous interval covers the beginning of next one. This approach is based on the fact, that initial conditions have significant effect on wave parameters only during the first day of computation, and on barotropic currents and sea level only during the first several days. But atmospheric effect remains dominating factor during the whole period of computation.
Ensemble prediction method is perspective one for grid-technology implementation. Ensemble of wave and current fields is determined on the basis of ensemble of wind fields, obtained from atmospheric models. Prediction which is result of ensemble averaging provides more reliable results in comparison with prediction computed by using only one wind field. As a rule one can say that the more dimension of ensemble the better quality of prediction. It is evidently the ensemble prediction method can easily be implemented in grid, because computations for each ensemble member are completely independent.
Functioning of Wind Wave and Current Models in the Grid Environment ARC
ARC software has the user commands which provides uploading of tasks, checking of their status, receiving of results. Program text in the xRSL language (Extended Resource Specification Language) is a list of pairs of elements (Attribute = «Value»). The simplest executing of one copy of model in ARC occurs the following way: xRSL-script is formed, which defines the name and location of executing program, its arguments, lists of input and output data, file names of standard output flow and errors; program is uploaded into the grid-environment by command ngsub and task execution is monitored of by command ngstat; as the results of task execution are ready they are downloaded from grid by command ngget.
In application to sea dynamic problems under consideration it is necessary for each copy of the model to make the following procedures: to form initialization file with appropriate parameters; to generate xRSL-script; to upload the task into grid; to download the results from grid. In case of large amount of input data gradations the algorithm automation is needed. With these object the special program solutions SGM (SWAN Grid Manager) and PGM (POM Grid Manager) were developed in MHI which general algorithm may be formulated the following way. The first stage is creation of set of initialization files (for each gradation of wind velocity and direction in case of multiversion calculation or for each start and finish date in case of retrospective analysis). On the next stage for each initialization file corresponding xRSL-script is generated and sent into the grid-queue. After execution of all tasks the downloading and processing of simulation results are occurred. The possibility of task monitoring and deleting from grid-queue is provided as well as detailed logging of executing operation.
Conclusion
Such problems of sea dynamics as multiversion calculation of wind waves and currents, their retrospective analysis and ensemble prediction can be effectively solved in the grid environment.
The special software SGM (SWAN Grid Manager) and PGM (POM Grid Manager) is developed in MHI to support functioning in the Nordugrid-ARC environment of wind wave model and barotropic current model correspondingly. This software arranges setting of user parameters, forming of configuration files, generation of xRSL-scripts and their execution in the grid-environment, checking of formed grid-task status, receiving of their execution results, cancellation of executing grid-tasks and restart of program.
Texts of PGM and SGM scripts with test examples one can found on the following web-pages:
–
fomin.hpc-mhi.org/public/grid/gpom– program codes and test example of solving the problem of wind current re-analysis in the Sea of Azov;
–
fomin.hpc-mhi.org/public/grid/gswan1 – program codes and test example of solving the problem of wind wave multiversion calculation in the Sea of Azov;
–
fomin.hpc-mhi.org/public/grid/gswan2– program codes and test example of solving the problem of wind wave re-analysis in the Sea of Azov.
Within the described frameworks of interests we are looking for cooperation and discussion with DRIHM members and partners.
