DATA_DISTRIBUTION (.param)
Keyword type
String
Description
This keyword determines the parallelization strategy used. Available options are:
- Kpoint - only k-point parallelization will be used (best scalability)
- Gvector - only g-vector parallelization will be used (worst scalability)
- Mixed - a combination of k-point and g-vector parallelization will be used
- Default - the optimal parallelization strategy for the architecture will be used
The Default setting is appropriate in most situations.
The Kpoint option is available only when the number of processors is a factor of the number of k-points.
The Gvector option is appropriate for calculations involving large systems where often only one k-point is used.
The Mixed option is available when the number of processors and the number of k-points have a common factor (for example 6 k-points and 4 processors).
The differences in the scaling properties of the different schemes dictate the number of processors that should be used. For example, a calculation that uses 16 k-points is likely to finish faster on 4 processors (using Kpoint distribution) than on 6 processors (using Mixed distribution) and most likely faster than on 9 processors (using Gvector distribution).
Default
Default
Example
DATA_DISTRIBUTION : Gvector