Next: Specific parameters for the
Up: Parallel version of ALIAS-Maple
Previous: Parallel procedures in Maple
Contents
Before using the parallel implementation it is compulsory to set some
variables:
- `ALIAS/bisection_master`: the number of
unprocessed boxes
that the master will have to generate before dispatching the boxes
to the slaves (default value: 30). It is also the bound on
the number of
unprocessed boxes that the master will compute when all the
slaves are busy before checking again if any slave is free (within the
master algorithm there is a mechanism to monitor if a slave has
completed its calculation but stopping the master algorithm allows to
ensure that all the computation is not done by the master). This
number should be small so that the master is not doing any heavy
computation while the slaves are waiting for boxes. This variable
corresponds to the C++ variable ALIAS_Parallel_Max_Bisection
- `ALIAS/working_directory`: the full path of the directory
in which the C++ master and slave programs will be
created
- `ALIAS/pvm`: the path of the include files for the pvm
library e.g. "/u/caphorn/PVM/pvm3/include"
- `ALIAS/libpvm`, `ALIAS/libpvm_linux`: the path of
the pvm library respectively for SUN and LINUX PC's. You should have
typically something looking like
"/u/caphorn/PVM/pvm3/lib/SUN4SOL2" and "/u/caphorn/PVM/pvm3/lib/LINUX"
- `ALIAS/gpp_sun`, `ALIAS/make_sun`: the full path
for the C++ compiler and the make program for SUN
workstation. Typically "/usr/local/bin/g++" and "/usr/ccs/bin/make"
- `ALIAS/gpp_linux`, `ALIAS/make_linux`: same for
PC's under linux.
Hence a typical Maple file for solving a system of equations EQ in the variable VAR within the ranges S on a mixture
of SUN and PC will be:
`ALIAS/profilN`:="/net/coprin/intervalles/Profil-linux":
`ALIAS/libN`:="/net/coprin/intervalles/Lib-linux":
`ALIAS/working_directory`:="/u/Interval/Maple":
`ALIAS/bisection_master`:=2:
`ALIAS/diam_switch`:=0.3:
`ALIAS/bisection_slave`:=15:
ParallelHessianSolve(ef2,VAR,S,["otway","molotova"],[[30,4000],[30,4000]]);
As soon as the Maple procedure run two C++ program will be generated:
one contained the keyword MASTER, the other one with the keyword
SLAVE. The program MASTER will be compiled on the machine
that run Maple and will control the process. The program that will run
the slaves is the SLAVE program: one or two such programs will
be generated according to the architecture of the slave computers
(slave program running on a Sun workstation will have their name ended
by SUN while the program running under Linux will be ended by
LINUX).
Next: Specific parameters for the
Up: Parallel version of ALIAS-Maple
Previous: Parallel procedures in Maple
Contents
Jean-Pierre Merlet
2012-12-20