./README_SMP
*
* README for IBM Visualization Data Explorer for Symetric Multiprocessors
*
*
*
The information contained in this file consists of hints for using the
Data Explorer SMP for symmetric multiprocessors.
========================================================================
CONTENTS
To go directly to an item, search for the item number beginning
at the left margin.
1. The Partition module
2. Partitioning an already-partitioned field
3. The default amount of memory used
4. The default number of processors used
5. The difference between SMP and distributed
========================================================================
1. The Partition module
With Data Explorer SMP, the Partition module should be used to
create a composite field from the imported data.
Only if the data set is partitioned
will Data Explorer SMP exploit the multiple processors available.
Note that you can run partitioned data equally well on both
uniprocessor and multiprocessor machines.
2. Partitioning an already-partitioned field
Once a field is partitioned it will not be repartitioned if it
is passed through the Partition module again. Thus to maximize
the use of the available processors, you may want
to increase the number of partitions (the "n" paramter for
Partition) from the default value if you will be operating on
only a subset of the data later on. Note however, that over-partitioning
a field leads to extra overhead.
3. The default amount of memory used
By default on a MP machine, less memory will be used than on
a UP machine. This is under the assumption that other users
may be more likely to be using an MP machine. The amount of
memory used can be changed with the -memory command line option
to dx. Usage("memory",0) will show you how much memory is being
used.
4. The default number of processors used
For an MP machine, if the number of processors (n) is 4 or greater,
the default number of processors used is n/2, rounded down. If the
number of processors is 2 or 3, then the default number of processors
used is 2. The number of processors used can be changed with the
-processors command line option to dx.
5. The difference between SMP and distributed
When you run parallel on an SMP machine, intramodule parallelism
is used. For example, the Isosurface module will operate in
parallel on different portions of a partitioned field. When
you use the distributed method of execution, different modules
run on different machines; for example, Isosurface might be
running on one machine while MapToPlane is running on another.
There is overhead involved with using distributed computation, as
the data must be transferred between machines.
========================================================================
The following trademarks apply to this information:
IBM is a registered trademark of International Business Machines Corporation.
IBM Visualization Data Explorer is a registered trademark of International
Business Machines Corporation