yes, thanks!
" manalyzer has automatic per-module multithreading "for free" "
does that mean that if one needs modules to run sequentially (think first clustering, then making higher level physics objects), the user needs to enforce that?
> > reworking our old (MuCap inherited) mana.cpp
> > To have a good start, what is advised:
> > a) following the Rootana Analyzer Framework as advertised in https://midas.triumf.ca/MidasWiki/index.php/Rootana_Analyzer_Framework with the TRootanaEventLoop
> > b) The manalyzer advertised in https://bitbucket.org/tmidas/rootana/src/master/manalyzer/
> > Option b) with it's module support?
>
> It depends on your experiment and your data. If your data is just a few numbers,
> either analyzer would do (even the old midas mana.c analyzer would do).
>
> But if you have complicated data, especially streaming data, where midas events
> and physics events do not have a 1-to-1 connection, the old analyzers will
> not cut the mustard. I have done this kind of processing in the old analyzers,
> came up with a generalized structure, the "m" analyzer is the result.
>
> Historical note: I developed rootana maybe 15 years ago to process data
> from the ALPHA experiment at CERN (anti-hydrogen trap), more recently, I developed
> manalyzer to process more complicated data from the ALPHA-g experiment at CERN (vertical
> anti-hydrogen trap, to study gravity effects). Joseph McKenna from CERN contributed
> the automatic by-module multithreading and benchmarking.
>
> To compare:
>
> - both rootana and manalyzer use the same TMidasOnline class for connecting to live MIDAS data
> - both use the same midasio classes to read and write MIDAS files
> - rootana uses the older TMidasEvent class to access MIDAS banks
> - manalyzer uses the newer TMEvent class. the main improvement is that TMEvent is
> a well-behaved C++ class with a usable assignment and copy-constructor operators.
> - graphics and web interface uses the same ROOT graphics, ROOT web server (a derivative
> of mongoose web server library) and the same JS-ROOT.
> - the manalyzer has better defined control flow and C++ object lifetime rules
> wrt run start and run end (this corrects problems with life time of C++ objects
> in the old rootana).
> - manalyzer has automatic per-module multithreading "for free" (thanks, Joseph!),
> I typically see the alphag analyzer use 8-10 CPU threads without having to use
> any threading or mutex gunk in the analysis code.
>
> The current plan is to include manalyzer into MIDAS as a git submodule, with luck
> it will happen for the next midas release, the hold up is with the midasio class that needs
> to be updated, git-submoduled and added to midas. Also the old TMidasOnline class needs
> updating.
>
> To conclude, we use manalyzer for ALPHA-2 and ALPHA-g, people who use it seem to be
> generally happy (especially with the automatic per-module multithreading), we use
> it for a couple of simpler experiments at TRIUMF.
>
> I hope this answers your questions?
>
> K.O. |