Page 57 - Data Science Algorithms in a Week
P. 57

Using Deep Learning to Configure Parallel Distributed Discrete-Event Simulators  41

                           Complexity Parameters that Capture the hardware/software Structure of a Parallel
                                              Distributed Discrete-Event Simulator
                        Number of Global Nodes                                                  4
                        Mean Local Nodes per Computer                                           1
                        STD Local Nodes per Computer                                            0
                        Mean Number of cores                                                    1
                        STD Number of cores                                                     0
                        Mean processor Speed                                                   2.1
                        STD processor Speed                                                    0.5
                        Mean RAM                                                               6.5
                        STD RAM                                                                1.9
                        Critical Path%                                                        0.32
                        Theoretical Speedup                                                     3
                        Local Events/(Local Events + External Events)                           1
                        Subscribers/(Publishers + Subscribers)                                 0.5
                        Block or Scatter?                                                       1

                        Table 4. TW has the minimum wall clock time for the aircraft detection problem
                                      using 4 Global Nodes and 1 Local Node with Block


                                     Time Management and   Best (Minimum Wall Clock Time)
                                    Synchronization Scheme
                                    BTW                                  0
                                    BTB                                  0
                                    TW                                   1


                       Methodology

                          This is the methodology devised in order to recognize the best time management and
                       synchronization scheme for a PDDES problem. The input vector is define based on the
                       complexity  and  features  of  the  software,  hardware,  and  messaging  of  the  PDDES
                       problem (as explained above). The output vector defines the best time management and
                       synchronization scheme (TW, BTW, BTB). This pattern matching is achieved using a
                       DBN  trained  with  case  studies  performed  by  a  Parallel  Distributed  Discrete-Event
                       Simulator. This methodology is depicted in Figure 9.
   52   53   54   55   56   57   58   59   60   61   62