I am running a problem with a sliding interface in parallel. On 2 processors it takes about twice as long to run as on a single processor. I have : turned off hyperthreading removed monitors reduced file outputs to a minimum streamlined the udf used for the motion

I tried adding in each component of the model, turbulence, sources, ideal gas, material properties and the moving mesh where the interfaces do not communicate.

The slow down occurs with the moving mesh and a sliding interface are activated.

I have tried partitioning along the interface, perpendicular to the interface, encapuslating, non encapsulated, segregated and coupled solvers.

FLUENT believes it is a communication slow down.

I have tried on windows and LINUX platforms without any change.

Are there any workarounds for this problem? Has any one overcome this problem?

If you are using the sliding- or dynamic mesh - method the grid rotation for every timestep always takes place on the host computing node. Therefore you have large communication traffic between your computing nodes and this slows down your simulation.

The best speedup i have reached on a 4-cpu shared memory workstation was 2.4 to 2.6 for a simulation using sliding mesh . On our cluster the speedup was almost lower because we have only gigabit-ethernet connection between the nodes.