[Paraview] Questions on MPI

[Moreland, Kenneth]

Chewping,

I recommend reading the section on "Visualizing Large Models" in our
Supercomputing 2008 tutorial handouts (available from
http://www.paraview.org/Wiki/SC08_ParaView_Tutorial), or better yet attend
the tutorial in a few weeks.

I also provide terse answers to your questions below.


> 1.      How MPI distribute the work load?

The reader units each read in part of the data in the file(s).

> 2.      Does the master node divide the work load evenly first, then
> distribute to each slave node?

The master node never touches the data.

> 3.      Or the master node pass the whole chunk of work load to every slave
> node, then the slave node OEtake¹ their own piece of work to process?

The server nodes (what you call slaves) each take their own piece by
directly reading it from the file.

> 4.      Can the work load being distributed according to priority (load
> balancing)? Or give a bigger portion of work load to the better performance
> node?

There is no facility for that right now in ParaView.  In my experience, the
benefits you gain are seldom worth the effort in either development or added
overhead in computing and redistributing the data balance.  You are welcome
to try though.

-Ken


   ****      Kenneth Moreland
    ***      Sandia National Laboratories
***********
*** *** ***  email: kmorel at sandia.gov
**  ***  **  phone: (505) 844-8919
    ***      fax:   (505) 845-0833