[R] R killed on Solaris

Ott Toomet otoomet at econ.dk
Wed Jan 14 17:49:40 CET 2004


thanks for you reply.  Actually I think it has more to do with solaris
than with R.  The simple workaround is to run the program

 | Date: Wed, 14 Jan 2004 13:24:02 +0100
 | From: Uwe Ligges <ligges at statistik.uni-dortmund.de>
 | 200M of the data size processed or 200M of RAM consumption by R?
 | Does the .Rout file tell you anything interesting?

This is 200M in all.  I dont think it is quotas, perhaps something
with stack size, but unfortunately I dont understand the manual
(e.g. http://docs.sun.com/db/doc/806-1075/6jacsnimq?a=view):



   A process, which attempts to allocate large amounts of memory
   either as an array or by using malloc, fails when launched by the
   shell. This problem has been seen while allocating 240,000,000
   elements as either an array of doubles or using malloc to allocate
   the 1,920,000,000 bytes of space.


   This can have one of two causes. Resolve it accordingly.

   1. Lack of swap space

   Try running the program as root on the console; if it runs, this is
   not the problem.

   2. Stack size and data segment size are in conflict

   If  the stack size is set too large, this can conflict with the
   data segment, and the process cannot
   be started. Setting the stack size to the default value of 8192
   resolves this problem and allows the
   programs to start.

I have used stack size 8192, ulimits seem to be ok.

 | 200M RAM is nothing, hence only an issue if there are quotas defined on 
 | your system.
 | Uwe Ligges
 | BTW: R-1.8.1 is recent.

R 1.7.1 may still be the most recent piece of software on that box.
You have probably seen servers before with strict usage policy and
admins who don't care much about software.  Other ,,cool'' stuff
include gcc 2.95 (which cannot make 64-bit executables on sparc) and
ghostscript 5.10.

Best wishes,


 | > Hello,
 | > 
 | > I am running a preliminary data-processing job on solaris.  
 | > writes the results into a dataframe for latter processing.
 | > 
 | > The problem is that when I run the job interactively (in emacs/ESS),
 | > everything works fine.  If I try to run it from shell using R CMD
 | > BATCH, R will be killed after around 100 000 lines.  Solaris says
 | > simply "killed" and thats it.  The total memory consumption seems not
 | > to be a problem, around 200M on a 4G machine.

More information about the R-help mailing list