[ut.dcs.graphics] Hierarchical Parallelism for an Illumination Alg. :Friday 9th

dunc@eecg.toronto.edu (Duncan Elliott) (03/02/90)

                Electrical Engineering Computer Group
                         Cider Seminar Series

     Hierarchical Parallelism for a Global Illumination Algorithm

                                  by
                           George Drettakis
                       Dynamic Graphics Project
                    Department of Computer Science
                        University of Toronto

         Time: Friday, Mar. 9, 1990, 12:05 --- Place: GB 220

       Global  illumination  algorithms  can  create  highly-realistic
  computer  generated  images. FIAT is a global illumination algorithm
  developed at the University of Toronto that uses an  adaptive  space
  subdivision  technique,  and  relies  on  discrete  approximation to
  simulate the distribution of light.  The algorithm is very expensive
  in  both  computation and memory.  In this talk we introduce a novel
  algorithm  that  utilizes  hierarchical   parallelism   to   improve
  execution speed and to meet the high memory demands.

       In the approach presented a two-level hierarchy of  parallelism
  is  used.   The first level assigns groups of subvolumes, created by
  space subdivision, to each station available  in  a  loosely-coupled
  environment.   Multiprocessor  workstations  allow a second level of
  parallelism by splitting the computation inside each subvolume  into
  tasks  which  are  assigned  to  individual  processors  within  the
  workstation.

       The design was implemented and tested on workstations connected
  by  a  local  area  network, resulting in substantial improvement of
  execution speed, and  satisfactory  distribution  of  memory  across
  stations.

       Amazing pictures and a 2-hour animation will be shown. :-)

                             Coming Soon

       Date         Who                     Topic

      Mar. 16   Garth Gibson   (2pm) Using Reliable Disk Arrays
                               To Avoid I/O Bottlenecks