The nsteps and the computing time of Peri3dController

Hello,
    I want to couple DEM and FEM in order to implement the multiscale analysis just like the method in Ref[1]. I choose the engine Peri3dController to do it, because this engine has many advantages, such as it can use strain to get stress directly and it can save the former strain state by using p3d.strain=strain. In the FEM, I use the central difference method to compute, and it has stability condition that the loading step is very small. The result is that the compute times are very big and each strain increment is very small.
    In the DEM, I called the engine Peri3dController each compute time and I found if I use the constant nSteps=4000, it will be a very big computation cost. So I want to find an appropriate nSteps that can guarantee the result is relative steady and meanwhile can be small to reduce the computation cost. I also find if I called the Peri3dController too many times, the compute time will increase along with each calling. I do not whether it is related to that I use saveTmp() and loadTmp().
    I accord to the file example/test/peri3dController_triaxialCompression.py, the file use two engine, can I use one engine and only change the goal of the Peri3dController to get the same result?
    Can you give me some suggestions? Thank you.

Keqi

[1] http://www.engmech.cz/2012/proceedings/pdf/018_Stransky_J-FT.pdf