Academia.eduAcademia.edu

Figure from:Parallelization of geant4 using top-c and marshalgen

See full PDFdownloadDownload figure
Figure 4. Number of aggregated tasks and running time for ParN02 for 50,000 events can be found in examples/novice/NO2. The simulations used Geant4 version 4.6.2 and gcc-2.95. They were run on dual processor Pentium machines with | GB of RAM running at 1.3 GHz on the lxplus cluster at CERN. The ma- chines were running RedHat 7.3. The machines were lightly loaded. The running time in each case excludes approxi- mately two minutes that Geant4 uses to load its libraries and initialize its data structures on all master and slave nodes.

Figure 4 Number of aggregated tasks and running time for ParN02 for 50,000 events can be found in examples/novice/NO2. The simulations used Geant4 version 4.6.2 and gcc-2.95. They were run on dual processor Pentium machines with | GB of RAM running at 1.3 GHz on the lxplus cluster at CERN. The ma- chines were running RedHat 7.3. The machines were lightly loaded. The running time in each case excludes approxi- mately two minutes that Geant4 uses to load its libraries and initialize its data structures on all master and slave nodes.

Related Figures (3)

Figure 1 shows an example of using Marshalgen annota- tions to generate marshaling routine for the class Foo. Figure 1. Marshalgen annotations for a sim- ple class
Figure 1 shows an example of using Marshalgen annota- tions to generate marshaling routine for the class Foo. Figure 1. Marshalgen annotations for a sim- ple class
Table 1. Optional Annotations: one of five de- fault cases, determined by parsing data types Some of the possible Marshalgen annotations are listed in Table 1. For more details on Marshalgen, see [7, 8].
Table 1. Optional Annotations: one of five de- fault cases, determined by parsing data types Some of the possible Marshalgen annotations are listed in Table 1. For more details on Marshalgen, see [7, 8].
Figure 3. The speedup of parallelized Geant4 applications. (DMX = Dark Matter Experi- ment) The communication overhead, although almost the same as MPI, does incur some cost. Therefore, the speedup de- pends on the granularity of each particular Geant4 applica- tion (see Figure 3). With 50 processes, the master is ob- served to use approximately 30% of the CPU time. With 100 processes, the master is observed to use approximately 50% of the CPU time.
Figure 3. The speedup of parallelized Geant4 applications. (DMX = Dark Matter Experi- ment) The communication overhead, although almost the same as MPI, does incur some cost. Therefore, the speedup de- pends on the granularity of each particular Geant4 applica- tion (see Figure 3). With 50 processes, the master is ob- served to use approximately 30% of the CPU time. With 100 processes, the master is observed to use approximately 50% of the CPU time.