Alvin,
I'm only posting these results to make a point. I made a simple plate with a hole, ran a mpa analysis with about 1000 shell elements, and it only took five seconds to run and obtain convergence within 1% (spa about two seconds). I've seen a lot of models where people try to break up the elements around a hole to make it look like a hand mesh, but with the p-element, you don't really have to do this. I created a lot of elements on purpose, and I certainly don't need this many to meet convergence. Try auto-meshing your model without Edge Distribution and check your results. Hope this helps.
