ucdavis.edu (10/29/90)
Question: Do numerical methods or algorithms which vectorize well tend not to parallelize well, and vice-versa? Background: We use mathematical models (mostly 1-D) to model flow and mass transport in an estuary. Currently the flow model uses an explicit Method-Of-Characteristics scheme which we wish to replace with some sort of implicit scheme (either MOC, Finite Element, or Finite Difference). Since we have a network of Sun sparcs we are considering the ability of the various algorithms to be parallelized. We are also considering buying multiple cpu machines (but not the ncube, massively parallel variety). We also want to consider the ability to vectorize, in case we run on a Cray or buy a vectorizing machine ourselves. Our current explicit algorithm doesn't vectorize at all, and is slow on a serial machine (because the time step has to be kept short). However, we think it would be suitable to parallelization; and also think that what makes it suitable to parallelize would be lost by going to an implicit scheme (which would vectorize better). But this is just speculation on our part, and we would like to hear others' thoughts on this. -- Ralph Finch 916-445-0088 rfinch@water.ca.gov ...ucbvax!ucdavis!caldwr!rfinch Any opinions expressed are my own; they do not represent the DWR