GIDE: Graphic Interface for Discrete Element - Visual Computing: Scientific Visualization and Imaging Systems - page 66

Step 1 Step 1 Step 2

Var 1 XX 1 Var 1 XX 1 XX 2

Var 2 YY 1 Var 2 YY 1 YY 2

... ... ... ... ...

Var N ZZ 1 Var N ZZ 1 ZZ 2

Fig. 3.5 Internal representation 2

Table 3.1 Results of the tests

A B C D E

Real 0 min 4.596 s 0 min 0.351 s 0 min 0.974 s 11 min 36.040 s 0 min 0.557 s

User 0 min 4.023 s 0 min 0.001 s 0 min 0.485 s 0 min 9.820 s 0 min 0.227 s

Sys 0 min 0.417 s 0 min 0.292 s 0 min 0.411 s 7 min 1.452 s 0 min 0.292 s

For a solution to the third point a series of tests on the various solutions had to
be devised, and the best solution selected. The possible solutions are as follows:

• Write the data in text form, using standard recording primitive (current

solution);
• Write the data in raw binary form, using the standard recording primitives
(reference solution);
• Use the HDF5 format of the first version of the GIDE;
• Use an initialised matrix at maximum size;
• Use a matrix of empty size, increasing it by one vector at each time step.

Each of these solutions was tested with the recording of 4,096 variables of the
long integer type (32 bits), over 4,096 time units. These tests were carried out with
the 'time' command under Linux, and the result is given in the Table 3.1 .
The good results from solution B are due to the possibility in binary mode of
recording a vector in a single operation, stating the memory address and the length
of this vector. However, this solution does not satisfy the prerequisites overall.
Aside from this point, it clearly appears that solution 'E' gives the best results,
and this is therefore the solution selected and integrated in the GIDE. Moreover, it
offers the advantage of satisfying our three initial constraints.

3.2.2 3D Motor

The choice of 3D library is crucial. Though this interface is considered as a
navigation tool in an ocean of data, the first contact with the user is nevertheless
visual and this is his first expectation: to display the results of the simulation.
There was no question of making an nth 3D library, but rather of choosing the
most appropriate one from existing ones. Four constraints were set: