[Rivet] [Rivet-svn] r2972 - trunk/data/anainfo

[Frank Siegert]

On 25/02/11 11:30, David Grellscheid wrote:
>> Is this anywhere close to an issue yet? I would have thought that the 
>> analyses are basically negligible in their memory usage, but would 
>> certainly like to see numbers for that.
> 
> I guess it's still very far from the 1 GB common blocks that LHAPDF
> assigns, but given that the list of analyses will grow monotonically,
> there needs to be some consistent scheme of where they get put once
> there's a thousand of them instead of 117.

So I tried to get the numbers for the memory usage:

$ python
Python 2.6.6 (r266:84292, Sep 15 2010, 15:52:39) 
[GCC 4.4.5] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import os
>>> ret=os.system("pmap %s" % os.getpid())
[...]
 total     9656K
>>> import rivet
>>> ret=os.system("pmap %s" % os.getpid())
[..., libRivet, fastjet, HepMC]
 total   214924K
>>> all_analyses = rivet.AnalysisLoader.analysisNames()
>>> ret=os.system("pmap %s" % os.getpid())
[..., libRivet, fastjet, HepMC, all analysis libraries, e.g.:]
00e35000    764K r-x--  /home/fsiegert/rivet/install/lib/RivetCDFAnalyses.so
00ef4000     12K r----  /home/fsiegert/rivet/install/lib/RivetCDFAnalyses.so
00ef7000      8K rw---  /home/fsiegert/rivet/install/lib/RivetCDFAnalyses.so
 total   219596K

So if my test case is correct, simply loading the analyses costs
us 5M for all of them. The biggest of them, CDF, is something like 800K.

So unless you find a flaw in the above (I had never used pmap before)
I'm not worried about this.

Cheers,
Frank

PS: The 215M for bare Rivet remain also if I remove all analysis libraries,
so nothing seems to be hidden in there.