Spalart-Allmaras instability
 

Ok, Im working on a test case with fluent 6.1, and have recently changed the mesh layout, resulting in much more cells picking up the turbulent flow in the wake of the body (but no change to Y+, or distribution "along" the flow, purely tangential to the surface). OK, the body has kelvin-helmhotz instability (well the wind tunnel does), and initial sims on the old meshes backed this up, they converged at around 30,000 iterations for a RANS solution.

However, with the new mesh, S-A is converging normally until around 10,000 iterations, then goes mad, with iterations picking up by around 2 or 3 orders of magnitude and CL and Cd going crazy. K-E just converges normally at around 30,000-50,000 for the same mesh. Can anyone shed any light on why the Spalart model is doing this?? (it is consistenly happening on different mesh densities and different body shapes, 25 and 35deg rear facing slope)