$Id: README,v 1.18 2011/02/05 18:40:10 johns Exp $

Notes on the Unix builds of Tachyon:
------------------------------------

  General suggestions on configuration:
  -------------------------------------
    When choosing which way to build Tachyon for a given platform,
  there are a few general suggestions which will give you good results.
  When running on a small shared memory machine, you'll probably get the 
  best performance from the multithreaded builds of Tachyon.  If you're 
  running on a large SMP system with 30 or more processors, you may
  get equivalent or possibly better performance by using MPI, depending 
  on the efficiency of a vendor's threads implementation.  

  If you're going to run Tachyon on a distributed memory cluster, then 
  you'll want to use one of the MPI or LAM builds options.  In my experience,
  LAM gives far better performance when running on a cluster connected by
  ethernet than MPICH running on the same hardware.  So, if you're not using
  a vendor-provided MPI tuned for a custom interconnect, I highly recommend
  using LAM as your first-best option.

  Configuration:
  --------------
    You may wish to enable support for various special features such
  as interactive spaceball fly-throughs, JPEG and PNG image format
  support, and other configurable features.  These are all controlled
  in the file Make-config.  Similarly, other options which affect
  operation of the ray tracer, such as floating point precision, 
  mailbox data structures, etc, are enabled or disabled by editing
  this file.  Make-config also includes paths for include and linkage
  paths for libraries like MPI, OpenGL, and other libraries.

  Compiling:
  ----------
    Simply type 'make' to see a list of build configurations, choose a 
  configuration and type "make xxxx-xxx-xxx" as appropriate.  The resulting 
  binaries will be built and linked in the associated 
  "../compile/xxxx-xxx-xxx" directory.

  Running Multithreaded Builds:
  -----------------------------
    Running the multithreaded builds of Tachyon is trivial, it is
  just like running any other program.  No special flags are required
  unless you want to limit the number of threads (and thus CPUs) used
  during the rendering process.  By default, Tachyon will attempt to 
  spawn a number of computation threads equal to the number of CPUs
  installed in the machine, as determined by operating-system-specific 
  code built into Tachyon.
   
    Autodetect number of threads/CPUs:
    ----------------------------------
    % cd ../compile/solaris-thr
    % tachyon ../../scenes/balls.dat

    Specify number of threads/CPUs (42 in this example):
    ----------------------------------
    % cd ../compile/solaris-thr
    % tachyon ../../scenes/balls.dat -numthreads 42

 
  Running MPI Builds:
  -------------------
    Since the details of using 'mpirun' and similar commands vary 
  substantially from one platform to another, and sometimes even between
  different site-specific installations here is a short list of typical
  commands to execute the parallel versions of Tachyon.
  Each of the examples below illustrate how to run Tachyon on the
  "balls.dat" or "tetra.dat" scene files included with the distribution.  
  The default output image is "outfile.tga" and will be written in the working
  directory unless you override this default with the 
  "-o /somedir/somefile.tga" flags to Tachyon.

    MPICH
    -----
      % cd ../compile/solaris-mpi
      % mpirun -np 16 tachyon ../../scenes/balls.dat

    LAM-MPI
    -------
      In order to run the regular LAM builds follow these steps
      % cd ../compile/linux-lam
      % lamboot 
      % mpirun -c2c -np 16 tachyon -- ../../scenes/balls.dat
      % wipe

      In order to run the OpenGL-enabled LAM builds, the DISPLAY 
      environment variable must be propagated to the running job, thus
      some extra parameters must be passed to LAM's mpirun which will make
      it inherit your existing DISPLAY environment variable:
      % cd ../compile/linux-lam
      % lamboot 
      % mpirun -c2c -x DISPLAY -np 16 tachyon -- ../../scenes/balls.dat
      % wipe

    Scyld (MPICH)
    ------------- 
      Scene files must be copied to a globally accessible file area, which
      may be an issue for some Scyld installations.  I'm working on making
      the root node re-distribute geometry to the rest of the nodes so that
      file I/O ability is needed only on the root node. 
      % cd ../compile/solaris-mpi
      % mpirun -np 16 tachyon ../../scenes/balls.dat

    IBM LoadLeveler
    ---------------
      A load leveler batch job script needs to be written, 
      a short example LL script is shown below:
      #@ class = short
      #@ job_type = parallel
      #@ node_usage = not_shared
      #@ environment = COPY_ALL
      #@ tasks_per_node = 32
      #@ node = 1
      #@ wall_clock_limit = 0:30:00
      #@ output = $(host).$(jobid).$(stepid).out
      #@ error = $(host).$(jobid).$(stepid).err
      #@ queue
      export MP_SHARED_MEMORY=yes  # use shared memory for MPI comm.
      poe /home/stonej/tachyon /home/stonej/balls.dat -o /tmp/outfile.tga

    Cray T3E
    --------
      % cd ../compile/cray-t3e-mpi
      % mpprun -n 12 tachyon ../../scenes/balls.dat

    Sun ClusterTools 4.0 MPI
    ------------------------
      % cd ../compile/solaris-hpc
      % mprun -np 0 ./tachyon ../../scenes/balls.dat

    Compaq Alphserver SC at PSC
    ---------------------------
      % rinfo
      % qsub -I -l rmsnodes=2:8
      % prun -N 2 -n 8 ./tachyon ../../scenes/balls.dat
      % ^D

    Intel iPSC/860
    --------------
      % cd ../compile/ipsc860-mpi
      % getcube -t16
      % load "tachyon ../../scenes/tetra.dat"  
      % relcube
 
    Intel Paragon XP/S
    ------------------
      % cd ../compile/paragon-mpi 
      % pexec "tachyon ../../scenes/tetra.dat" -sz 16

    Mercury RACE Multicomputers (ppc, i860, SHARC)
    ----------------------------------------------
      % cd ../compile/mercury-ppc-mpi
      % ln -s tachyon tachyon.ppc
      % mpirun -np 16 -h 0x800000 tachyon ../../scenes/tetra.dat
   
      Note: the -h flags explicitly set the heap size for the
            process.  Some scenes will require more heap size than
            others, so this flag may or may not be needed in order to
            increase the default heap size provided to Tachyon.

      For non-MPI builds on the Mercury machines, use the following
      sequence to initialize and load the code on a node, where XXX
      is the node you desire to run on:
        % sysmc -v -ce XXX init &
        % sleep 3
        % runmc -v -h 0x800000 -ce XXX tachyon.860 ../../scenes/tetra.dat

    CSPI Multicomputers (ppc)
    -------------------------
      % cd ../compile/cspi-ppc-mpi
      % mpirun -np 16 tachyon ../../scenes/tetra.dat


