[Rd] R 2.9.0 slower than R 2.8.1 for the data.frame function

Philippe Hupé Philippe.Hupe at curie.fr
Mon May 18 18:24:25 CEST 2009


Dear developers,


I have noticed difference in computation time for the data.frame
function between R2.9.0 and R2.8.1. The older release is more efficient:
typically, R2.9.0 spends three more time in the data.frame function.
Therefore, when many calls of this kind is done inside a function, the
additional time spent with R2.9.0 may be several minutes.

Looking at the profiling results (see file below), it seems that R2.9.0
spends time in nchar function while it was not the case with R2.8.1.

I am runing R under debian stable on 64-bits architecture (see the
cpuinfo file for details). Both R2.9.0 and R2.8.0 have been compiled
from source with the same default options.

The data I use can and the R script can be retrieved (see below) for
people who are interested in doing some benchmark.

I am looking forward to have your feedback for this problem.


Best,

Philippe


profiling with R2.8.1:
http://xfer.curie.fr/get/X3oSg7VO81F/profiling-2.8.1.txt

profiling with R2.9.0:
http://xfer.curie.fr/get/he4A8yQfe3j/profiling-2.9.0.txt

RData:
http://xfer.curie.fr/get/SOOaBe5nMz3/mydata.RData

R script:
http://xfer.curie.fr/get/w64axb61KPt/comparison-2.9.0-vs-2.8.1.R

computer information:
http://xfer.curie.fr/get/xRS9VfMqA5l/cpuinfo.txt

-- 
Philippe Hupé
Institut Curie, CNRS UMR 144, INSERM U900
26 rue d'Ulm
75005 Paris - France
 	
Email :  Philippe.Hupe at curie.fr
Tél :	 +33 (0)1 56 24 69 91
Fax:     +33 (0)1 56 24 69 11
website : http://bioinfo.curie.fr



More information about the R-devel mailing list