Improving Between-Shot Fusion Data Analysis with Parallel Structures Page: 3 of 10
This report is part of the collection entitled: Office of Scientific & Technical Information Technical Reports and was provided to Digital Library by the UNT Libraries Government Documents Department.
The following text was automatically extracted from the image on this page using optical character recognition software:
IF (myrank EQ root) THEN BEGIN
PRINT,'restoring original mdsvalue'
RESTORE, '/scr_storagel/veitzer/mdsplus/zipfitfile/txmdsproxy. sav'
ENDIF ELSE BEGIN
RESTORE, '/scr_storagel/veitzer/mdsplus/zipfitfile/txmdsclose. sav'
The implementation of this parallel MDSplus data handler was tested by retrieving CERQUICK
data from an MDSplus server. We show in figure 3 timing information comparing time to retrieve
the data both by making straight MDSplus calls (the situation in figure 1) and by using the
proxy server (the situation in figure 2) in figure 3. One can see in the figure the server bottleneck
as a function of increasing number of processors trying to simultaneously access the data (pink
points). Using the MPI/MDSplus proxy server to distribute the data shows an improvement
over making direct MDSplus calls (blue points), although a bottleneck is still seen because the
underlying MPI communication protocols used in this case were blocking. The red points in
figure 3 show the mean time to make a MDSplus call for the proxy server. This shows the proxy
server is efficient, because the time to retrieve data from the MDSplus server does not increase
with the number of processors. Also, when the proxy server is used, there is only one connection
to the MDSplus server made, which allows other researchers to also access the MDSplus server
in a timely fashion. These tests were done on the Tech-X Athlon cluster with up to 15 working
processors. We also tested the proxy server on a GA cluster with similar results.
Demonstrate parallelization of fusion data analysis codes
Central to doing real-time fusion data analysis is the ability to do a number of basic data analysis
functions quickly. For instance, FFTs are basic parts of many fusion data analysis routines. In
the case that the FFTs are done on data that is independent, for instance on time-series' which
are taken at different spatial locations in the tokamak, a task-based scheme for parallelization
can be employed. In the Phase I project we demonstrated that using a task-based parallelization
was an effective way to gain speedup in the calculation of toroidal mode power spectra at the
NSTX tokamak. A main advantage to this work is that parallelization of the existing IDL code
requires no knowledge of parallelization techniques, only knowledge of IDL syntax and language
The data analysis path for calculating mode power involves the following steps. There are
twelve Mirnov channels spread out around the tokamak in the toroidal direction. During each
shot each channel measures magnetic fluctuation data at that location. During the analysis
of the data, a FFT of each channel is made to measure the power spectrum at that location.
Then for each frequency, a thresholding method is used to determine if a toroidal mode at that
frequency is present. The method of determining if a mode is present requires data from each
FFT. The number of modes that can be measured is limited by aliasing in the toroidal direction.
Typically this data analysis is slow for a couple of reasons. The time to retrieve the Mirnov
Here’s what’s next.
This report can be searched. Note: Results may vary based on the legibility of text within the document.
Tools / Downloads
Get a copy of this page or view the extracted text.
Citing and Sharing
Basic information for referencing this web page. We also provide extended guidance on usage rights, references, copying or embedding.
Reference the current page of this Report.
NIETER, CHET. Improving Between-Shot Fusion Data Analysis with Parallel Structures, report, July 27, 2005; United States. (digital.library.unt.edu/ark:/67531/metadc787371/m1/3/: accessed September 20, 2017), University of North Texas Libraries, Digital Library, digital.library.unt.edu; crediting UNT Libraries Government Documents Department.