Problems with running time

Hi, I wonder if your performance measures are valid. When you think of applying the results to programming a modell, you would have to initialize and define local pointers each time, which is time consuming for shure.
So when you do a performance measure for a method do you include all the prerequistes in the loop or not?

Hi Nico,

I believe your comments apply mostly to my comparison of math expressions vs. comparetext(), I was simply trying to isolate the equivalent of a comparetext() in math terms and not include speed difference between declaring doubles and strings. Again, I encourage you to perform the test you would like to see and post results/script.

It is also true that my test involving gettablenum()s, getlabelnum(), global variables and arrays is not an equivalent battle. I believe the value is in knowing the relative speeds of these commands in the event you have a choice of how you setup your model. For example, maybe you have a few select global parameters in your model, and you want to keep them in a Global Table but they get used a lot in various triggers in your model. Based on the speed results, you may consider creating corresponding Global Variables and setting them to the table values on a reset event. Then use the Global Variables in your model logic instead of gettablenum().

I am not sure whether Cliff declared the LocalVariablePointer before every excecution of the get... in his test. Yes, it would be interesting to see the difference. I think in many cases (like a schedule) within a given trigger you might read an entire row of data or create items/events based on numerous rows (like a day's worth of rows). In these cases the number of times you have to declare the LocalVariablePointer start to decrease significantly.

Kris

Hi,
I am currently developping a new Flexsim library which will enable me to simulated continuous processes. I need to model chemical reactors, crushers and particle separators. I had a hard time trying with the Fluid Library and I needed too many customized blocks.
To produce this new library, I wrote hundreds of lines code. It is design to be very fast if running in Flexscript, but I wanted to have the possibility to switch to C++ and compile easily. It effectively compiles without a single warning.

After a few months of development, I wanted to know if all that pain was worth something. I benchmarked an old study we have realized of a continuous process that was modeled with the Discrete objects only. This process was re-modeled but with the new library. The simulation contains 150 blocks and spreads over a full year of operations including reliability, shcedules and tasks sequences. Hardware used: Dell Precision M70 (2GHz, 2G of RAM), WinXP SP3 and Flexsim 4.52. Here are my findings:
1) The old simulation was 110 times slower (1400 seconds versus 13 seconds) than the version built with my library. This little 13 seconds also provides a complete mass balance and the handling of several chemical species and particle sizes.

2) Flexscript vs C++: numberous simulations provided a nice rule of thumb : C++ is 2.4 times faster than Flexscript (in version 4.52). The model is almost pure code with no events nor flowitems.
I found really interesting that 2.4 factor between C++ and Flexscript. For small models, it means to me that I don't need to care about compiling. For large models, compiling will be required only for repeated runs such as during optimization.

Question: what will this factor become in Flexsim 5.0? I will be very pleased if I discover it under 2...

Vincent Béchard, ing., M.Sc.A.
Simulation & optimization
__________________________
SNC-Lavalin Devonyx Inc.
360, St-Jacques O.
Montréal (Canada) H2Y 1P5
514-845-2166 #2204

I had same problems some time ago. Didn´t think about change query. I believe the easiest way to speed up the modellrun very much, is to set the STOP TIME and close all view windows. After the run stops I open the relevant windows and see what data has done.