About | Formal basis | Workpackages | Documents | Meetings | Partners | Contact list |

 »   Workpackages

 »   WP 8 Parallel visualization for large-scale computations

Leader »
Prof. Jacek Kitowski
ACK CYFRONET AGH, Kraków
Co-executor » TASK
Start date » 6 Jan, 2003
Ending date » 31 Oct, 2004
[ Timetable |  Team ]
Short task description

One of the main areas of large-scale computations are problems resolved on a computational grid using the finite element method or finite difference method. Because large tasks are highly time-consuming, even when run parallel, it is crucial to be able to watch the results online. The visualization itself requires a significant amount of time. In connection with the above the need of possessing efficient tools for parallel visualization becomes extremely important.

Research will concentrate on the following areas:

  • Parallel visualization portal
    Definition of functionality requirements, technology and architecture of the portal and its subsequent implementation; definition of visualization functions on the computations and portal side; creating a module for multimedia streams management.

  • Parallel visualization of computational program results
    Specification of functionality requirements of the parallel visualization system, architecture designing and implementation of the modules: preparing data for visualization, parallel processing and rendering with the help of the pV3 library.

  • Using the CUMULVS library to synchronize work of parallel computational programs

The pV3 (parallel Visual3) library is a graphic library designed in the Aeronautics and Astronautics Department on the Massachusetts Institute of Technology and intended to parallel 3D visualization. The library has been designed specially for visualization of the data generated in a distributed system, assuring the maximum possible independence of executing solver calculations.

The PVM library is proposed as the software for management of data transmission between network nodes. The library provides all necessary features, including efficient data transfer, message transmission and indication of a specific host for executing the given part of a job. pV3 may also be used together with the MPI library. In such case the MPI interface may be used for parallel solver activity management and PVM sends the data for visualization. Such work sharing results from the fact that pV3 uses numerous PVM features that have not been included to MPI.

An extension to the solution of collecting and sending data is using the CUMULVS library from the Oak Ridge National Laboratory together with a separate visualization system (for instance AVS, Open Data Explorer or another one). CUMULVS is a peculiar kind of interface that makes it possible to comfortably fetch data from the current parallel application and send them to the visualization toolkit. It also allows for controlling the values of the user parameters in a distributed simulation program and interactive choice of scale and amount of data that should be visualized. Thanks to that it is possible to visualize large amounts of data without overloading the network with unnecessary transmissions.