CTH variable memory use is inefficient and wrong
This issue was created automatically from an original Mantis Issue. Further discussion may take place here.
ParaView 3.12.0, remote server (8 processes), Linux client.
-
Start ParaView. Open Dave's large CTH program (256 files).
-
Turn off "compute derived variables".
-
Apply (note that we have not added any variables)
-
Surface. (No idea why this seems to matter, 30 mbytes will be used, each process).
-
ssh into the node we have. Do "top". Note the memory for each pvserver, as well as the total. I get the following: 2.19 GBytes used total, RES of 135 MBytes for each of the 8 pvservers.
-
Note that the Information tab states that 3.6 MBytes are in use. There are 3.5 million cells and 4.7 million points.
-
So far, so good.
-
On the Properties tab, load Material Volume Fraction 1.
-
Information tab states that we are using 7.2 MBytes.
-
Top states that 2.32 GBytes total is used on the node - an increase of 130 MBytes of memory use! Each process has grown to 151 MBytes, thus about 15*8=120 MBytes is in use! In theory, the Information tab is correct - we should have malloced about 3.5 million bytes.
This is killing very large visualizations on Cielo. It needs to be fixed ASAP.