Jayesh,<div> Thanks for getting back to me. I'm not using MPI-IO functions. My datasets are unstructured grids and I have to use another toolkit (VTK) to read the files. I can store them in the binary format and they shouldn't be more than a few megabytes. </div>
<div> My MPI processes are launched after i read the data and every processor requires knowledge of the complete dataset (in other words i cannot partition the data and split it across the available processors). Hope I answered your question correctly.</div>
<div> So, given this should I let all the nodes read the input files? </div><div>Thanks</div><div>Prashanth<br><br><div class="gmail_quote">On Tue, Dec 7, 2010 at 10:45 AM, Jayesh Krishna <span dir="ltr"><<a href="mailto:jayesh@mcs.anl.gov">jayesh@mcs.anl.gov</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">Hi,<br>
This will depend on how much data is read and where your MPI processes are launched. Are you using MPI-IO functions (like MPI_File_read_all()) for reading the data ?<br>
<br>
Regards,<br>
<font color="#888888">Jayesh<br>
</font><div><div></div><div class="h5">----- Original Message -----<br>
From: Prashanth <<a href="mailto:prashanth.dumpuri@gmail.com">prashanth.dumpuri@gmail.com</a>><br>
To: <a href="mailto:mpich-discuss@mcs.anl.gov">mpich-discuss@mcs.anl.gov</a><br>
Sent: Mon, 06 Dec 2010 10:44:53 -0600 (CST)<br>
Subject: [mpich-discuss] reading data into a parallel program<br>
<br>
All,<br>
I have a main program that reads input data from files and 3<br>
filters/classes instantiated within the main program that process the<br>
datasets. I have parallelized 2 of those 3 classes (using MPICH2). Is it<br>
faster to read data on the root processor and then broadcast it to all the<br>
child nodes or should I let all the nodes read the data?<br>
I'm running my main program on a Windows XP machine with an Intel Quad<br>
core processor.<br>
Thanks<br>
Prashanth<br>
<br>
</div></div></blockquote></div><br></div>