[petsc-users] superlinear scale-up with hypre

[Barry Smith]

    Cannot really say without more information about what is taking  
time on 32 cores and 256 cores.

    If you run 32 core and 256 core with -log_summary (also --with- 
debugging=0 ./configure version of PETSc) we'll be able to see where  
the time is being spent and so if it makes sense.

    Barry

On Mar 8, 2010, at 1:09 PM, Christian Klettner wrote:

> Dear PETSc,
> I am using a fractional step method to solve the Navier-Stokes  
> equation
> which is composed of three steps. I have to solve a Poisson equation  
> for
> pressure in Step 2 and I use the GMRES solver with Hypre's BoomerAMG  
> for
> preconditioning. I have tested for strong scaling using a fixed  
> problem
> size of 16million degrees of freedom and varied the number of cores  
> from
> 32 to 256. I have found superlinear speedup up until this number of  
> cores.
> Is there a reason why BoomerAMG exhibits this kind of behaviour?
>
> Best regards,
> Christian
>