[R-sig-ME] Blue model timing out.

[Sean MacEachern]

Hi,

I�m having troubles with memory allocation when running a BLUE with approximately 9000 genotypes using LMER or ASREML. I�ve looked at increasing the workspace in ASREML to over 12 gb, and from what I can see I�m only using ~6gb. However, I constantly get errors about insufficient workspace. Conversely, if I fit an animal model with the 9000 genotypes plus 3000 additional parents I can obtain BLUPs without too much trouble. Has anyone experienced any problems fitting BLUE models with reasonably large numbers of genotypes?



One thought I have would be to use tryCatch, and withTimeout to cancel the BLUE model if it didn�t run within a certain time frame, but this doesn�t work with ASREML or LMER.

As an example:



system.time(tryCatch({ asr_blu <- withTimeout( {areml(data=bdat, fixed = Trait~Genotype, random= ~Field+Rep)}, timeout=5,elapsed = 5)}, TimeoutException = function(ex) cat("BLUE Timed out\n")))



BLUE Timed out

user system elapsed

1462.257 10.843 1484.458





The timing shows that CPU and elapsed time is going way longer than the 5 seconds I�ve specified in the withTimeout call. Does the native code in asreml-r not check for user interrupts, which I believe is needed to interrupt the function? Is there another way to kill a function if it goes into a zombie state?



Regards,



Sean MacEachern


	[[alternative HTML version deleted]]