[Beowulf] problem : mpi dynamic scheduling ??

[Greg Lindahl]

On Tue, Jul 05, 2005 at 01:04:03PM -0400, Mark Hahn wrote:

> in other words, large-tight apps are incompatible with meaningful
> load-balancing.  

Mark,

There are a couple of examples of tightly coupled apps that do either
static or dynamic load balancing, but in most cases this load
balancing was programmed in by the programmer above the MPI level. For
example, ocean codes don't have to compute on land, so you can static
load balance by using different-sized rectangles for your data
decomposition. A system that does this invisibly is Charm++, used by
namd, a molecular dynamics code. In namd's case the workload at a
given point varies over time.

Another classic example which isn't solved is in weather forecasting.
Air columns with cumulus action (convection) take more time to compute
than air columns without. But the ratio of computation to data size
is small enough that it would require a really good interconnect to
enable load-balancing this part of the computation.

This is the kind of stuff that you'd never expect Mosix to magically
load-balance.

-- greg