<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
</head>
<body>
<div>I have a great idea:</div>
<div>A small group should get together to discuss the parallel interface.</div>
<div>Cheers</div>
<div>Andreas </div>
<div><br>
</div>
<div><br>
</div>
<div><br>
</div>
<div id="composer_signature">
<div style="font-size:85%;color:#575757" dir="auto">Sent from my Samsung Galaxy smartphone.</div>
</div>
<div><br>
</div>
<div style="font-size:100%;color:#000000"><!-- originalMessage -->
<div>-------- Original message --------</div>
<div>From: Martin Nolte <nolte@mathematik.uni-freiburg.de> </div>
<div>Date: 26/07/2017 08:28 (GMT+00:00) </div>
<div>To: dune-devel@lists.dune-project.org </div>
<div>Subject: Re: [Dune-devel] On which side is DataHandleIF::size() called </div>
<div><br>
</div>
</div>
<font size="2"><span style="font-size:10pt;">
<div class="PlainText">Hi Markus,<br>
<br>
you assumption seems correct to me: In general, the receiver cannot know how <br>
much data was sent. Loadbalancing for a discrete function space with <br>
element-dependent polynomial degree would be a simple example.<br>
<br>
Unfortunately, there are many situations in which the receiver does actually <br>
know how much data is sent (e.g., Finite Volume, i.e., 1 DoF per element), but <br>
there is no way to communicate this to the grid.<br>
<br>
There are further deficiencies of the current interface in more complex <br>
scenarios (e.g., how do you sent data of different type?) that cannot currently <br>
be handled. Two years ago, we started discussing over a new concept for a data <br>
handle. Maybe it is finally time to make it happen.<br>
<br>
Best,<br>
<br>
Martin<br>
<br>
On 07/26/2017 09:18 AM, Markus Blatt wrote:<br>
> Hi,<br>
> <br>
> when I first had to deal with parallelizing a DUNE grid I somehow<br>
> assumed that the size method is only allowed to be called on the<br>
> sending side. Why else what the scatter method have the size as an<br>
> additional parameter.<br>
> <br>
> Somehow I was not sure whether that assumption really holds or every<br>
> grid developers thinks so. I just wrote a handle to be used for<br>
> loadbalancing a grid from a master process to all. As a precaution I<br>
> introduced an assertion in the size method that makes sure it is only<br>
> called on rank 0. For UGGrid this assertion just failed.<br>
> <br>
> Is my assumption wrong, is it a bug in UG or am I missing something<br>
> fundamental here?<br>
> <br>
> Cheers,<br>
> <br>
> Markus<br>
> <br>
<br>
-- <br>
Dr. Martin Nolte <nolte@mathematik.uni-freiburg.de><br>
<br>
Universität Freiburg phone: +49-761-203-5630<br>
Abteilung für angewandte Mathematik fax: +49-761-203-5632<br>
Hermann-Herder-Straße 10<br>
79104 Freiburg, Germany<br>
<br>
_______________________________________________<br>
Dune-devel mailing list<br>
Dune-devel@lists.dune-project.org<br>
<a href="http://lists.dune-project.org/mailman/listinfo/dune-devel" target="_BLANK">http://lists.dune-project.org/mailman/listinfo/dune-devel</a></div>
</span></font>
</body>
</html>