I'm at a workshop right now, and my webmail client is "unpleasant", so I'll 
reserve a more complete answer until later, but I can confirm your results.

I've run examples.phase.anisotropy  under a couple of different configurations 
on my 10.5.8 MacBook Pro.

At 100x100, PySparse runs 100 steps in 14 s, compared to 37 s for Trilinos. 
(I'm on the battery, so these are both slower than this machine is capable of).

At 200x200, PySparse takes 47 s (only 3x as long for 4x the number of cells), 
but Trilinos takes 182 s. Almost 5x longer than the smaller grid and about 4x 
slower than PySparse! So the performance gets worse as the problem gets bigger 
(neither process is memory-bound on this machine).

I don't know what the issue is, nor do I know what solver is being used. We'll 
have to look at this in more detail. We may be able to make smarter default 
choices.

In general, I'd say that Trilinos *can* be beneficial, with proper choice of 
solvers and preconditioners. In some cases, PySparse cannot solve equations at 
all that Trilinos can handle easily. Still, it is clearly not a blanket 
superior choice.


> P.S. Just curious:  would it be difficult to measure the time the tests
> spend in the critical portions of the tests?  

Yes.


Reply via email to