Hi Jed,<br><br>I've a better understanding now. I think I can use it for my poisson eqn solving which has pressure at the cell center. However, my U and V are staggered. May I right to say that I can still use DA to help the domain spilting process even for this scenario? However, can i solve my momentum eqn with DA still since there's staggering?<br>
<br>Thanks alot!<br><br><div class="gmail_quote">On Fri, Dec 11, 2009 at 7:00 PM, Jed Brown <span dir="ltr"><<a href="mailto:jed@59a2.org">jed@59a2.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
<div class="im">On Fri, 11 Dec 2009 18:24:58 +0800, Wee-Beng Tay <<a href="mailto:zonexo@gmail.com">zonexo@gmail.com</a>> wrote:<br>
<br>
> But you mention abt latency, so shouldn't minimizing the number of neighbor<br>
> processes reduce latency and improve performance?<br>
<br>
</div>Perhaps, depending on your network. But there are many tricks to hide<br>
latency of ghost updates, global reductions (in dot products) are<br>
harder especially since MPI collectives are synchronous. The higher<br>
iteration counts are way more painful than marginally higher update<br>
cost.<br>
<div class="im"><br>
> For both do u mean dividing 1 big grid into 4 55x35 grids?<br>
<br>
</div>Yes, instead of 4 thin slices. And so on as you refine. DA does this<br>
automatically, just don't choose a prime number of processes (because<br>
then it would be forced into doing slices).<br>
<div class="im"><br>
> so whichever method I use (horizontal or vertical) doesn't matter? But<br>
> splitting to 4 55x35 grids will be better?<br>
<br>
</div>Trying to send directly from some contiguous array is not a worthwhile<br>
optimization. My comment about latency was to guide against another<br>
"optimization" of sending some components of a vector problem separately<br>
when not all components "need" updating (it's likely faster to do one<br>
update of 5 values per ghost node than to do two separate updates of 1<br>
value per node).<br>
<br>
Splitting into 4 subdomains isn't "better" than 2 subdomains, but when<br>
using many subdomains, they should not be thin slices.<br>
<br>
DA manages all of this. If you have some compelling reason *not* to use<br>
DA, you won't go far wrong by copying the design decisions in DA.<br>
<font color="#888888"><br>
<br>
Jed<br>
</font></blockquote></div><br>