On Sat, Nov 2, 2013 at 12:42 PM, John Peterson <jwpeterson@gmail.com> wrote:



On Sat, Nov 2, 2013 at 8:21 AM, Kirk, Benjamin (JSC-EG311) <benjamin.kirk@nasa.gov> wrote:
On Nov 2, 2013, at 9:03 AM, John Peterson <jwpeterson@gmail.com>
 wrote:

> Damn... Yeah this would clearly help the most for the parallel runs.  No reason to really optimize anything else until that's implemented!

So that is done now in the branch...

Nice work!  We are now well below 7 Gb now:


but the best part is it should scale much better while adding processors (doing some more runs to test that theory at the moment.)

Scalability of memory looks pretty good:

https://drive.google.com/file/d/0B9BK7pg8se_iM1hEYk9UcFY4WnM/edit?usp=sharing

The change in _total_ memory consumption with increasing cores is:

2 -> 3: +2.67 Gb
3 -> 4: +2.71 Gb
4 -> 5: +2.26 Gb

Since each copy of this mesh takes up about 2.2 Gb, these numbers are scaling almost perfectly.  My machine started swapping on 4 cores last time, now it can run on 5 cores with no problems.

--
John