memory issues with cssolver

27 views
Skip to first unread message

rharris

unread,
Oct 18, 2011, 9:29:59 AM10/18/11
to mddnmr
Hi All

I have begun using CS within mddnmr2.0 and am having some
interesting(!) memory usages.
For triple resonance backbone experiments, when running the
region.runs file, whilst the CPUs are running at 100% the amount of
RAM memory usage is negligible (~0.2%). For an aromatic 13C NOESY
where I have nimax=160 ni2max=32, the %Mem goes to 6.7% (I have 12Gb
of memory, so using about 1Gb), then for a HCCH-TOCSY nimax=96
ni2max=128 the %MEM goes up to 33.9% (so using just over 4Gb RAM per
calculation). I have seen this on two separate computers. Is this
behaviour normal?

thanks
Richard

Krzysztof Kazimierczuk

unread,
Oct 19, 2011, 9:47:31 AM10/19/11
to mddnmr
Dear all,

Below I paste whole discussion with Richard. Accidentally we discussed
the topic privately. Sorry for that:Hi Richard,

Ratio between amounts of memory used is more or less fine then. The
amount of RAM is ca. proportional NIMAXxNI.
The ratio between aroCnoesy and HCCHTOCSY is: (128x32x39x32)/
(96x128x30x128)=0.1
So, in this crude approximation Noesy should require ten times less
RAM then TOCSY. For HNCA it is again five times less.

Thank you for your report,

Krzysztof


W dniu 18.10.2011 23:04, Richard Harris pisze:
> Hi Krzysztof
>
> Thanks for the reply. Can I say CS is great! I have been using MDDnmr for a while but now even for the triple resonance experiments the difference with CS is very impressive and more so with NOESY/TOCSY spectra.
>
> In all cases so far I am limiting myself to 25-30% sampling
> So for HNCA: NIMAX 64 32 1 NI 512 1 1 (this was run under Bruker Topspin 2.1 hence the unusual definition of NI)
> The aroCnoesy: NIMAX 128 32 1 NI 39 32 1
> The HCCHTOCSY: NIMAX 96 128 1 NI 30 128 1
>
> hope that helps and thanks
> Richard
>
> On Tue, Oct 18, 2011 at 3:30 PM, Krzysztof Kazimierczuk <kk...@chem.uw.edu.pl> wrote:
>
> Hi Richard,
>
> Firstly, as an author of CS part of mddnmr I have to admit, that 3D CS algorithm is not well optimised yet. There are probably some non-optimal solutions in it, I have to find and clean them.
> 2D algorithm is much better tested. Thus, I appreciate your report very much and ask others to report bugs&strange issues.
>
> Memory requirements depend not only on nimax but also on ni. Could you mention the ni values?
>
> Thanks,
> Krzysztof


Reply all
Reply to author
Forward
0 new messages