Dear All,
I am writing on behalf of a user of our local cluster.
The OpenMX install is fairly recent (just with source('https://openmx.ssri.psu.edu/getOpenMx.R' in R). The script in question is rather long and starts with loading OpenMX:
require(OpenMx)
and subsequently snowfall:
library(snowfall)
sfInit(parallel = T, cpus = 64)
sfLibrary(OpenMx)
Apparently, OpenMx is recognized correctly as snowfalls 'sfLibrary' - during the setup all cores are utilised. When fitting models a CPU usage > 100 % is observed, albeit not all cores are used in average (as the individual fits are short). However, hitting a mxRun function step takes several days to execute.
The model in question is setup like:
HetModel <- mxModel("Heterogeneity Model",
+ funMG)
where 'funMG' is a mxFitFunctionMultigroup-object.
printing HetModel gives:
MxModel 'Heterogeneity Model'
type : default
$matrices : NULL
$algebras : NULL
$constraints : NULL
$intervals : NULL
$latentVars : none
$manifestVars : none
$data : NULL
$submodels : 'ModelForSample1', 'ModelForSample2', ...
$expectation : NULL
$fitfunction : MxFitFunctionMultigroup
$compute : NULL
$independent : FALSE
$options :
$output : FALSE
with a total of at least 200 submodels (hence the '...')
Is there any way to speed up the execution of mxRun? E.g. by dividing the task? (I tried setting useSocket = T, without effect).
Any pointer is appreciated.
Thanks a lot,
Christian