Hi,

I have a workstation hp xw8200 (xeon 3.2 GHz) with 1GB of ram windows XP Professional SP2 and fluent 6.1

first of all, I do my calculation in serial method the estimated time was 2 month (not possible). When I do the same calculation with parallel method, I have a estimated time to 15 days is better than the serial method but not enough.

But my question is about the processor's use. In serial method they are used at 50%-53% and in parallel method they are only used at 20%-23%. I think I forgot something when I prepare my calculation but I don't know what!! If you have some solutions to grow up the processor's use, I will be very satisfied.

thanks in advance.

sam

Turn off hyperthreading.

thanks for this information but I already turn off hyperthreading.

Strange. I run double 3 ghz Intel Xeon with 3gb of ram and XP Pro. When I use parallel processing, processor usage is of 100% and I really cannot do anything else on the machine. I used automated grid partitioning, did not change anything myself since it was the first time I did it. Maybe your grid is badly partitioned.

I don't know if my grid is badly partitioned, but I have meshed with a quad model (4 points) under Gambit.

I calculate a resolution of temperature and velocity. I choose for solver the segregated method, unsteady and I keep the rest by default. and I take turbulent model (viscous=>k-epsilon equation) for velocity.

Sounds like you are working with a relatively large number of nodes. Have you checked that the problem actually fits in the amount of RAM on your system? Things will of course be much slower if you are swapping in and out of RAM.

If I understand, you think I don't have enough RAM in my system. But when I read others message on forum, I see in generally configuration with 3GB of RAM but for calculation with very large number of nodes (1 million), and me I have only 8500 nodes.

If I understand the better scheme for a configuration of a computer is:

processor(s) <==> large amount of RAM to stock the model during the calculation.

and not:

processor(s) <==> RAM <==> HDD (virtual memory)

well on a three gb machine, i have used 2 million cells and 3 million cells (if all of them are tetra), i think the origin of you problem is the turbulence model you are using. now the question is how this can effect the calculations;

it can affect it in this way, usually when you use turbulence model, fluent choses flexible cycle as multigrid options for k and ep. Now in the flex cycle, you visit different grid levels based on the covergence you get on each grid level, further if you do not get convergence on one level it will move to next coarse level,

to add to this, you will notice in multigrid cycle, that there is one more paramter , called max iterations, which is set to 30, that is if you do not get converegnce needed in 2 iterations (the default value), fluent can iterate till 30 iterations (if the convergence is not reached).

in all the time for each iteration is not fixed and is variable, and for a difficult modell like k-e it may be very large.

so to test this thing, you can do this:

first of all change the turbulence model to something simple, like LES and then run and see if the calculation time is varying the same way.

second, chose V cylce with turbulence quantities and fix the max iterations to small value like 5, and then see how much time have improved.

sorry i think i did not read your messages properly, in your case the problem is your case is very small 8500 nodes is too small to run with parallel and get speed up, you get the advantage of parallel processing only when you have large cells or nodes sizes.

You say, it is better for me to calculate with serial processing than parallel processing when we have small number of nodes.

Ok thanks for yours answer.

and good luck in your work.

sam