Wayne Boucher wrote:
>Hello,
>
>Are you on a Linux box? And which version of Analysis are you using?
>(One nasty and one not-so-nasty memory leak was sorted in 1.0.13.)
>
>In theory a native OpenGL ought to be faster. (Brian could give you
>better chapter and verse on that one.) But 256 MB ram is definitely
>pushing it.
>
>As of 1.0.13 you can use pre-calculated contours (after you pick your
>peaks!). This should help a lot, especially on a low-memory machine.
>(But of course you cannot then more the contour levels up and down.)
>
>There is a -m option when you start up analysis. The default is 128 Mb.
>But that is only a measure of (most of) the C world memory, it excludes
>the Python world memory. So for you, "analysis -m 64" might be a better
>option. (Hard to call, that one.)
>
>
>
>
Hi Wayne
with respect to the -m option should you increase -m as you main memory size goes up?
another thing to say in passing is that part of the speed problem is with the cpu and io demands of contouring which are both high (is the contouring code threaded and will two processors help?) and part with multiple updates.
For instance
1. when i have a window with many small panes open such as a matches window for LinkSequentialSpinSystems I find that as I increase the number of match strips to say 15 the whole thing can get incredibly slow as it draws all the panes 3 times (I can point to what is going on in the code and have some reasonable ideas of how to debug the code but haven't had time to go further) I also note that one of the comments is 'interesting' # often fails first time... in updateMatches in LineSequentialSpinSystems:
if positions:
displayStrips(self.guiParent, positions, orthoPositions=None, spectrum=None, window=self.matchWindow)
# often fails first time...
self.update_idletasks()
displayStrips(self.guiParent, positions, orthoPositions=None, spectrum=None, window=self.matchWindow)
2. when scrolling throught complete sets of 3d spectra i.e. scan through all the planes of hnca and hncoca spectra ovelaid analysis can get slow. I guess this is because data on disk has to be paged into memory. One answer
I had for this which was to but yet more memory and store myspectrum data on a ram disk to which data would get copied to on startup....
as a general comment at Leeds we don't have any machines with less that 1G of ram. Our general mantra is that ram is cheap but postdoc/ researcher time time is expensive ;-)
Finally I would concur Brian Smith's comments from later on in this thread 'Let's give them credit for their huge achievement in
producing such a feature-rich programme in such a short time with so
few resources'
regards
gary
--
-------------------------------------------------------------------
Dr Gary Thompson
Astbury Centre for Structural Molecular Biology,
University of Leeds, Astbury Building,
Leeds, LS2 9JT, West-Yorkshire, UK Tel. +44-113-3433024
email: [log in to unmask] Fax +44-113-2331407
-------------------------------------------------------------------
|