This page guides you through using parametric distribution with Flux.
Introduction
This chapter discusses the establishment of a parametric distribution in Flux on a
single machine. The distributed computing allows the user to save
computation time while distributing several independent configurations of a same
Flux project. For example a Magnetic Transient project may be distributed regarding
different values of parameters such as the geometrical parameters (size, shape
etc...) or physical parameters (supply current values, speed...) varying in the
scenario. Several projects will run at the same time simulating all those
configurations, the main parameter of a distributed computation is the following
one:
The number of cores (i.e the number of running Flux in parallel);
The following topics are covered in this documentation:
Principles;
How to set up a parametric distribution on a single machine;
The parametric distribution allows the user to save computation time by parallelizing
several independent configurations of a finite element problem with different values
of I/O or geometric parameters varying in the scenario instead of running
them sequentially. To set-up a distributed computation, a primary Flux project is
mandatory in order to compute all its sub-projects having an independent
configuration.
The primary Flux project controls all the others secondaries projects (distribution)
and is in charge of the gathering of all the results obtained during the solving
process by all the sub-projects Flux as depicted in Figure 1 below:
Figure 1. Picture illustrating a parametric distribution, a primary Flux is
controlling several secondary Flux with independent configurations of
geometric or I/O parameters.
For this configuration, each secondary Flux has the same properties:
They all ran with the same number of cores which has been configured in
Supervisor Options;
The amount of memory assigned to the secondary Flux is set to the same value
as the primary Flux, if the memory is setted to Dynamic for the primary
Flux, the secondary Flux will also start with dynamic memory.
How to set up a parametric distribution on a single machine
In Flux 2D and in Flux 3D, the parametric distribution options (number
of secondary Flux in parallel) may be setted in the Supervisor's options or by
clicking on the button Distribution manager in the bottom right of the
Supervisor. For the Flux Skew module, this feature is not yet available and
will run automatically in sequential mode.
In any case, the parametric distribution may be setted as follows:
The Distribution manager may be started in the Supervisor's options,
in the Parallel Computing menu, in the section Distributed
Computing and then Parametric Distribution click on the
button Set local ressources as shown below;
Figure 2. The supervisor's options allowing the user to set the parametric
distribution.
An additional window Distribution manager should appears and is
asking to allow some resources (Number of secondary Flux in parallel) for
the parametric distribution, click on Allow;
The Distribution manager then asks for the Number of concurrent
Flux as follows:
Figure 3. The distribution manager asking for the Number of Concurrent
Flux.
This number is directy linked to the number of cores available on the
machine, the numbers of varying parameters are automatically distributed
over the number of concurrent Flux. The secondary Flux are automatically
run in monocore mode but may be changed by using the slider of
Number of cores per concurrent Flux.
Finally, click on Use to complete the parametric distribution
The parametric distribution may be started by clicking the button
Distribution manager in the bottom right of the Supervisor as
shown below.
Figure 4. The distribution manager button in the Flux Supervisor.
Define a parametrized scenario in Flux
Once that the distribution has been setted in the Supervisor, it may be employed as
follow in a Flux scenario:
Check the box Parametric distribution in the selected
Scenario.
Note: All the varying parameters
should be declared as Controlled parameters in the Control
parameters tab of the scenario as depicted below in Figure 5.
Figure 5. Scenario GUI box, highlightning in (1) the Parametric distribution that
is enabled and in (2) a list a of several varying parameters.
Example of application
To show the interest of the parametric distribution, let us consider a project
modeling a three-phase, eight-pole permanent magnet synchronous machine (PMSM) using
a Flux 2D Transient Magnetic application. This simulation will be controlled by the
angular position of the rotor from 0 to 90 degrees with imposed speed which is a
time dependant scenario. During the parametric distribution Flux will compute the
results for all the parameters combination for each time step.
Figure 6. Three-phase, eight-pole permanent magnet synchronous machine (PMSM)
described in Flux 2D.
The goal is to do a parametric distribution over two parameters:
The speed which is declared as an I/O parameter controlled by the scenario
and that is used by the rotating mechanical set
The shape of the magnet with the magnet outer arc value α setted with
a geometrical parameter as depicted below.
Figure 7. Magnet outer arc parametrized with a geometrical parameter that
may be selected as a varying parameter during the scenario.
Both parameters may have an influence on the performance of the electrical
machine. A table summarizing all the parameters is available below:
Table 1. Table summarizing the parameters, and their variation range
Magnet outer arc α (degrees)
Speed (rpm)
Minimum value
130
1300
Maximum value
170
1700
Step value
10
100
According to the previous, table the number of steps to solve is about 2525
(5*5*101) with five values for the speed, five values for the magnet outer arc over
a scenario with 101 time steps.
Note: Be aware that the time
steps cannot be separated in concurrent Flux, a strong time relation between the
steps is required to solve a Transient Magnetic application. This relation does
not exist in the Magneto Static application.
The results yielded by differents types of distribution using a different Number
of concurrent Flux setted in the Distribution manager are plotted in
Figure 8 while solving the same
scenario with 2525 time steps with a different value of Number of concurrent
Flux. The computation time with only 1 concurrent Flux (sequential
computing) is considered as the reference and is setted to 100% of the solving
time.
Figure 8. Graph representing the time computation evolution in function of the
number of concurrent Flux.
