[daip] making msclean run faster
Eric Greisen
egreisen at nrao.edu
Mon Mar 9 15:16:00 EDT 2009
Deidre Hunter wrote:
> Eric or other AIPS supporters,
>
> I am running multi-scale clean in IMAGR on a data cube of 2048x2048
> images of a big galaxy. My test of 6 channels is still running after
> 10 hours, so a full cube will take many CPU days. I am wondering if
> AIPS is using all it can of the computer's 5 GB of memory. My activity
> monitor suggests that there are frequent disk IOs and that only 2.3 GB
> of memory are in use. Is there anything I can set in AIPS that might
> allow IMAGR to use more memory and less disk IO?
>
> I am running 31DEC08, updated with MNJs, on a 2 x 3 GHz dual-core
> intel MacPro with MacOSX 10.5.6 and 5GB memory. The 31DEC08 was a
> binary installation.
One think to look at is the verb SETMAXAP. You might benefit from a
larger pseudo-AP. But the real questions are whether 2048x2048 is
appropriate, which resolutions are you using, and whether the I/O is
happening each time a channel starts up or at other times. The I/O is
likely to be slow when reading the full data set to strip out the
desired channel for imaging. If the I/O remains slow when using the
work file which is a minimal version of that channel's UV data, then
either you have a large number of points or your OS is unable to keep
the work file in RAM for some reason (too many other active processes,
some system parameter that limits the ram used in this way, ...). You
can look at the work file (class IMAGR) to see how big it is. Perhaps
you should be using multiple facets for 3D imaging issues and also for
having each image fit fully in RAM while you use it. SETFC will tell
you about the former if you ask it nicely (IMSIZE=0, CELLS=0).
I have worked hard to have IMAGR use more RAM when possible, doing as
much and as many facets as possible together in memory at once. When I
first saw your subject line I wondered if it is making images of too
many facets at each cycle (another capability I added). I just added in
31DEC09 yet another adverb to limit that in some cases.
Eric Greisen
More information about the Daip
mailing list