<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<style>body{font-family:Helvetica,Arial;font-size:13px}</style>
</head>
<body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;">
<div>
<blockquote type="cite" class="clean_bq" style="font-family: Helvetica, Arial; font-size: 13px; font-style: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: normal; orphans: auto; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: auto; word-spacing: 0px; -webkit-text-stroke-width: 0px;">
<span>
<div>
<div>Why not to use the original multipole solver? It is not really any less<span class="Apple-converted-space"> </span><br>
accurate.<span class="Apple-converted-space"> </span></div>
</div>
</span></blockquote>
</div>
<p>Oh? Then I guess we shouldn’t have bothered to write this whole paper about how much more accurate (and efficient) it is: <a href="http://adsabs.harvard.edu/abs/2013ApJ...778..181C">http://adsabs.harvard.edu/abs/2013ApJ...778..181C</a>. And going back
to the old solver doesn’t address Rahul’s problem at all. </p>
<p>Rahul, in the reference paper, we go to very high L_max, but in 2D only. Perhaps try setting mpole_3DAxisymmetry to .TRUE. in your .par file. This is likely not an ultimate fix since you need non-axisymmetric gravity, I assume, for your problem but it
could tell you if it is indeed the memory requirements of the mpole arrays.</p>
<p><br>
</p>
<p>Sean</p>
<p><br>
</p>
<div>
<blockquote type="cite" class="clean_bq" style="font-family: Helvetica, Arial; font-size: 13px; font-style: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: normal; orphans: auto; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; widows: auto; word-spacing: 0px; -webkit-text-stroke-width: 0px;">
<span>
<div>
<div>--<span class="Apple-converted-space"> </span><br>
On 08/11/16 12:41, Klaus Weide wrote:<span class="Apple-converted-space"> </span><br>
> On Wed, 10 Aug 2016, Rahul Kashyap wrote:<span class="Apple-converted-space"> </span><br>
><span class="Apple-converted-space"> </span><br>
>> Yes, I forgot to mention that I'm using new multipole implementation with<span class="Apple-converted-space"> </span><br>
>> 60 poles.<span class="Apple-converted-space"> </span><br>
>><span class="Apple-converted-space"> </span><br>
>> I have attached a small txt files with short summary on three runs which<span class="Apple-converted-space"> </span><br>
>> very well describes my problem. 1024 proc have been used for all runs with<span class="Apple-converted-space"> </span><br>
>> fixed lrefinemax and base blocks. I get three differenet error for three<span class="Apple-converted-space"> </span><br>
>> different maxblocks value.<span class="Apple-converted-space"> </span><br>
>><span class="Apple-converted-space"> </span><br>
>> My understanding was that reasonable use of maxblocks avoids any such<span class="Apple-converted-space"> </span><br>
>> memory failures.<span class="Apple-converted-space"> </span><br>
> Rahul,<span class="Apple-converted-space"> </span><br>
><span class="Apple-converted-space"> </span><br>
> It appears that the total memory required by<span class="Apple-converted-space"> </span><br>
><span class="Apple-converted-space"> </span><br>
> PARAMESH Grid + multipole solver ( + Particles + other units )<span class="Apple-converted-space"> </span><br>
><span class="Apple-converted-space"> </span><br>
> is just too much; I suspect that this is PRIMARILY due to the memory<span class="Apple-converted-space"> </span><br>
> requirements of the Multipole(_new) solver.<span class="Apple-converted-space"> </span><br>
><span class="Apple-converted-space"> </span><br>
> There are several large arrays allocated, see in particular statements<span class="Apple-converted-space"> </span><br>
> like<span class="Apple-converted-space"> </span><br>
><span class="Apple-converted-space"> </span><br>
> allocate (gr_mpoleScratch (1:gr_mpoleMaxLM,1:gr_mpoleMaxQ ),...)<span class="Apple-converted-space"> </span><br>
><span class="Apple-converted-space"> </span><br>
> in gr_mpoleAllocateRadialArrays.F90, where gr_mpoleMaxQ may be very large<span class="Apple-converted-space"> </span><br>
> and gr_mpoleMaxLM ~ gr_mpoleMaxL ** 2 ( == 60**2 in your case?).<span class="Apple-converted-space"> </span><br>
><span class="Apple-converted-space"> </span><br>
> Unfortunately this memory is required for each process, you cannot reduce<span class="Apple-converted-space"> </span><br>
> this by running on more procs.<span class="Apple-converted-space"> </span><br>
><span class="Apple-converted-space"> </span><br>
> It makes sense in general to try to reduce the memory required by the<span class="Apple-converted-space"> </span><br>
> PARAMESH Grid by lowering maxblocks, but this can go only so far;<span class="Apple-converted-space"> </span><br>
> maxblocks has to leave room for "a few" more blocks than the number<span class="Apple-converted-space"> </span><br>
> actually required by the distributed grid. These additional slots<span class="Apple-converted-space"> </span><br>
> are needed for temporary storage during processing by some internal<span class="Apple-converted-space"> </span><br>
> PARAMESH routines for things like block redistribution. I don't<span class="Apple-converted-space"> </span><br>
> know of a reliable way to predict how low "a few" can go in a given<span class="Apple-converted-space"> </span><br>
> case, so this has to be determined empirically. Apparently,<span class="Apple-converted-space"> </span><br>
> this was too low in your maxblocks=5 case.<span class="Apple-converted-space"> </span><br>
><span class="Apple-converted-space"> </span><br>
> It may be possible to tweak the maxblocks value further, possibly in<span class="Apple-converted-space"> </span><br>
> connection with also modifying the values of maxblocks_alloc and<span class="Apple-converted-space"> </span><br>
> maxblocks_tr (see amr_initialize.F90 and paramesh_dimensions.F90),<span class="Apple-converted-space"> </span><br>
> in order to allow the Grid domain initialization to proceed with<span class="Apple-converted-space"> </span><br>
> maxblocks < 10; but this may then still not give you enough free memory<span class="Apple-converted-space"> </span><br>
> for the multipole solver (and particles, etc).<span class="Apple-converted-space"> </span><br>
><span class="Apple-converted-space"> </span><br>
> So you should investigate ways to lower the memory requirements of the<span class="Apple-converted-space"> </span><br>
> Poisson solver; you may have to lower the resolution (not sure which<span class="Apple-converted-space"> </span><br>
> runtime parameters to change), or perhaps use a different implementation.<span class="Apple-converted-space"> </span><br>
><span class="Apple-converted-space"> </span><br>
> Klaus<span class="Apple-converted-space"> </span><br>
<br>
</div>
</div>
</span></blockquote>
</div>
</body>
</html>