<div class="clearfix">
<div class="field-items">
<div class="field-item even"><span class="date-display-single" property="dc:date" datatype="xsd:dateTime" content="2012-08-24T00:00:00+02:00">24/08/2012</span></div>
</div>
</div>
<div class="clearfix">
<div class="field-items">
<div class="field-item even">
<p>Dear users,</p>
<p> </p>
<p>We would like to remind you that on FERMI machine there are minimum default settings for the number of nodes </p>
<p>allocated to a single run. </p>
<p>The minimum bgsize is 64: if you ask less than that, you will get 64 nodes in any case. If you ask more than 64 and less than 128, they will be <em>round up to 128.</em></p>
<p>Since each node corresponds to 16 processors, the <strong><em>minimum</em></strong> number of processors allocated for a single run will be 1024,</p>
<p>even if your runjob asks for less resources.</p>
<p>This means that for 1 hour real time, you will use <em>at least</em> 1024 hours of your budget.</p>
<p>Examples:</p>
<p><strong>----- example I</strong></p>
<p># @ bgsize=64</p>
<p>runjob --np 1024 --ranks-per-node 16 -exe yourcode.x </p>
<p><strong>---- = 64 nodes allocated, 1024 MPI tasks (16 MPI tasks per node), 1h real time = 1024h budget</strong></p>
<p><strong>----- example II</strong></p>
<p># @ bgsize=64</p>
<p>runjob --np 512 --ranks-per-node 8 -exe yourcode.x </p>
<p><strong>---- = 64 nodes allocated, 512 MPI tasks (8 MPI tasks per node), 1h real time = 1024h budget</strong></p>
<p><strong>----- example III</strong></p>
<p># @ bgsize=128</p>
<p>runjob --np 512 --ranks-per-node 4 -exe yourcode.x </p>
<p><strong>---- = 128 nodes allocated, 512 MPI tasks (4 MPI tasks per node), 1h real time = 2048h budget</strong></p>
<p><strong>----- example IV</strong></p>
<p># @ bgsize=64</p>
<p>runjob --np 32 --ranks-per-node 16 -exe yourcode.x </p>
<p><strong>---- = 64 nodes allocated, 32 MPI tasks (16 MPI tasks per node on 2 nodes, 62 nodes idle), 1h real time = 1024h budget</strong></p>
<p><strong>----- example V</strong></p>
<p># @ bgsize = 100</p>
<p>runjob --np 1600 --ranks-per-node 16 -exe yourcode.x </p>
<p><strong>---- = 128 nodes allocated, 1600 MPI tasks (16 MPI tasks per node on 100 nodes, 28 nodes idle), 1h real time = 2048h budget</strong></p>
<p><strong> ----- example VI</strong></p>
<p># @ bgsize = 25</p>
<p>runjob --np 25 --ranks-per-node 1 -exe yourcode.x </p>
<p><strong>---- = 64 nodes allocated, 25 MPI tasks (1 MPI task per node on 25 nodes, 39 nodes idle), 1h real time = 2048h budget</strong></p>
<p> </p>
<p> As you can understand, it is very easy to burn your budget! So, please BE CAREFUL before submitting jobs.</p>
<p>Cross check your inputs and your job scripts, more than once!</p>
<p>When the job is running, you can check detailed info about your job with the command:</p>
<p>$ llq -l jobID</p>
<p>where jobID is in the form fen03.1000.0</p>
<p>In particular you'll be notified about the bgsize you requested and the real bgsize allocated:</p>
<p>example:</p>
<p>BG Size Requested: 2012</p>
<p>BG Size Allocated: 2048</p>
<div> </div>
<div> </div>
<p>However, in case of doubt, please don't hesitate to contact <a href="mailto:superc@cineca.it">superc@cineca.it</a>.</p>
<p>Happy computing!</p>
<p> </p>
<p>HPC Support Team @ CINECA </p>
</div>
</div>
</div>
<p>-- To unsubscribe to HPC-new send a mail to <a href="mailto:listserv@list.cineca.it?subject=any&body=unsubscribe%20hpc-news">listserv@list.cineca.it</a>.<br />
-- For more information see documentation at <a href="http://www.hpc.cineca.it/content/stay-tuned">http://www.hpc.cineca.it/content/stay-tuned</a></p>