http://onlinelibrary.wiley.com/doi/10.1002/fld.3827/abstract
They do shift the point where strong scalability breaks down, which is notable, but their claim that "employing PETSc in its elemental form for high performance computing (HPC) applications would yield very poor scalability" seems exaggerated to put it politely. Is there anything we can take away from this paper? Regards, John
