Author: Daniel Diefenthaler
Title: Online Visualization of Data Distribution Patterns
Abstract:
The data locality is of high importance for efficient execution of computations. In the high
performance computing context this is directly dependant on the data distribution. In this
work the common practices of different libraries are presented. Special attention lies on
the library DASH. DASH follows the PGAS (partitioned global address space) approach.
By doing so the pattern of the distribution is in the focus. The algorithm performing the
communication is implicitly deducted from that.
The main focus of this work is the visualization of such distribution patterns in DASH.
This should be possible over an online form. Therefore a client server structure is suggested.
A user interface and transfer formats for the construction and representation of such DASH
patterns are defined. This allows the server component that is a DASH application to print
out the pattern visualization to any machine.