GamBet computing power unleashed

GamBet, our Monte Carlo code for X-ray science and technology, has always been ahead of competitors with regard to special features (I’ll mention some later). In the past other packages may have had more raw computing power. We advanced the number-crunching capability of GamBet last year when we introduced parallel processing (http://fieldp.com/myblog/2011/multicore-gambet/). With this feature, GamBet runs could utilize the full power of a single, multi-core machine.

We’ve taken another step forward to make GamBet the best program for power users. We now support distributed computing. Runs may be divided between an unlimited number of machines. Here’s how it works:

  1. The new GamBet Distributed-computing Extension (GDE) is an add-on package for GamBet or Xenos Pro. The package includes two programs, GB_SOW and GB_REAP.
  2. GB_SOW includes all technical features of GamBet. The program may be installed on any number of computers without a license requirement. GB_REAP is installed on a master computer with a GamBet license.
  3. The user sets up and tests a GamBet run using the standard programs. Then, the required input files (GamBet control script, mesh definition, electric or magnetic field files,…) are sent to the worker computers.
  4. GB_SOW is launched on each worker computer, either from a window or from the command line. Multiple instances of GB_SOW may run on multiprocessor machines.
  5. The GB_SOW run produces a single binary file with a unique identifying name that contains all information from the GamBet calculation (escape particles, dose, statistics,…). The user moves output files from the worker computers to a data folder on the master computer.
  6. GB_REAP identifies all binary input files in the specified directory and combines them to produce GamBet output files in standard format. These files may be analyzed with GBView2, GBView3 or GenDist. Escape particles are combined in a single escape file with appropriate weightings. Statistical and dose information is averaged.

In the procedure, the user’s only role is to transfer files to and from the worker computers. This leaves complete flexibility for configuring the distributed network and automating file transfers. Despite its simplicity, the GDE concept has several advantages:

  • Calculations are independent and need not be synchronized.
  • There is no overhead for parallel processing. Five quad-core computers can reduce the time to generate a required number of showers by a factor of 20.
  • If one machine fails during the computation, data from the other computers is still valid and useful.
  • It is easy to improve the accuracy of a calculation by adding more showers without starting from scratch. The user simply creates more worker-computer files. They are added to the master folder and the total collection is recombined with GB_REAP.
  • Communication between computers is solely through file transfers. Therefore, it is easy to carry out extended calculations at any locations via a company network or the Internet.

While we’re on the topic of GamBet, it’s useful to review some other features that differentiate it from other Monte Carlo codes:

  • 2D and 3D geometries are defined with powerful finite-element mesh generators rather than idiosyncratic solid-modeling algorithms. Advantages include the direct input of complex shapes fromĀ  SolidWorks, ProE and other packages as well as extensive visualization options. The conformal meshes provide close fits to material surfaces.
  • 2D or 3D calculated electric and magnetic field solutions may be imported for detailed electron dynamics.
  • GamBet incorporates the Penelope 2006 physics engine for high-accuracy interaction calculations, from GeV energies down to less than 100 eV.
  • Calculations of fields, electron beam dynamics and target heating are integrated with Monte Carlo simulations in Xenos, a comprehensive software suite for X-ray source development.
  • GamBet features extensive graphical and quantitative post-processing capabilities for dose distributions and particle statistics.

For more information on GDE, please see http://www.fieldp.com/gde.html.

Comments are closed.