Hello, all!

First of all, I wanted to say thanks for all the great help I've received off
this list... in response to both my own questions and others.  This list is an
invaluable tool, and I appreciate the time put in by so many of you to answer
everyone's questions.

I posted a message to this list a few months back (8/30/99) reporting an
apparent memory leak when running OpenDx binaries on an SGI machine.  A few
responses were made, but nothing conclusive.  I've been pulled away to another
project since then, but am now returning to use the tools again, and the
memory-sucking is as annoying as ever.  =)  Basically, I have a fairly large
network (10 pages) with most (95%) of the modules set to "No cache", and the
others set to "last result".  But when I run a stepped display using the
sequencer, dx uses more and more memory with each step (~400k !!) until it runs
 out of available memory, when it crashes.  Jeff Braun volunteered to run my
network on his SGIs, and saw the same problem.  When he ran it under DX3.1.4,
there was no problem, which suggests something in the OpenDx code.

Since my last message, I've sat down to look at the problem in more detail. 
The net will run with no memory leaks, as long as the image is not rendered
(using image).  If I just use a print output to drive the network, it runs
fine.  There are slight memory increases as the sequence progresses, due to
larger selected data sets, but whenever I return to display the first image in
the sequence, it uses the same amount of memory as originally.  With the image
tool, though, the "used memory" reported continually increases.  Even when
returning to the first step in the sequence.

I wrote up a simple test case that loads a basic data set and displays a singe
glyph in space for each step.  Again, most modules are set to "no cache".  And
since the image being displayed is essentially the same for each time step (a
single glyph-arrow in space), the memory used at each step should be mostly the
 same.  When run without the "image" module, this net also has no memory
problems.  But with the image module, it appears to have a consistant increase
in memory with each time step (368 bytes).  When I ran the network on my WinNT
OpenDx, I saw the same problem, with an increase of (416 bytes) each time step.
  Note that as before, returning to the first time step does not cause the
memory use to return to the original value, but to continue to increase.

I know that these memory increases are small, but my test network is very
simple.  On the real network, the increases are more like ~400k with each time
step, which can eat up swap space real quick.  From my perspective, this looks
like an OpenDx code memory leak in the image module.  If someone is willing to
follow up on my tests, I have attached the test network (and can mail you my
original network, if you want).  Or if you've got any suggestions, I'd love to
hear them!

Thanks,
  Jeremy Zoss
  Southwest Research Institute
  (210)522-3089
  [EMAIL PROTECTED]

attached files: memTest.net, memTest.cfg, memTestData.csv

Attachment: MEMTEST.CFG
Description: Binary data

Attachment: MEMTEST.NET
Description: Binary data

Attachment: MEMTESTDATA.CSV
Description: Binary data

Reply via email to