I don't have any experience with R but I was heavy into S (the S Plus
commercial version) for 2 years. If the memory management is similar, then
the experience you are having is normal. The calloc and free functions are
actually using the interpreter's version of those functions. If you've got
the RAM, then the interpreter never bothers to do a memory garbage cleanup.
My guess is that on the writers of R did two different implementations of
the garbage cleanup routine. If they made the (reasonable) assumption that
the i386 version was running on a computationally weak system (compared to a
Sparc) then they might have traded away available RAM for speed.
Obviously this is just my opinion but I know S for i386 had some very
interesting memory quirks and features. I doubt it is a feature of the OS
as Linux/Unix tends to be a better memory manager than Windoze. And 32-bit
vs. 64-bit should be irrelevant to this issue.
Where are you measuring the memory loss? Is it at the R command line or
with a standard Unix process utility?
Matthew Zaleski
> -----Original Message-----
> From: Christopher Quale [mailto:[EMAIL PROTECTED]]
> Sent: Thursday, July 06, 2000 12:17 PM
> To: [EMAIL PROTECTED]
> Subject: [expert] [OT] differences between sparc64 and i386
>
>
>
> Sorry to go off topic, but I would like to get the
> perspective of the very knowledgable people on
> this list.
>
> Background: I am writing a program using the "R"
> language (a high level statistical language, similar
> to the "S" language). I am able to dynamically load
> *.so.0 libraries (written in C) to do computationally
> intense tasks in R functions. I am doing all the
> memory management using calloc() and free() and am
> absolutely, 100% sure there are no memory leaks. The
> function that calls the C code is repeated many
> times over.
>
> Problem: When I run the program on an i386 linux
> machine, the memory aggregates as the number of times
> the function runs within the R program, until I exit
> the R interpreter, at which time all the memory is released.
> However, running the recompiled (for sparc64 linux)
> program on a recompiled (for sparc64 linux) version
> of R, there is no such memory aggregation (or at least
> it is minimal compared to i386).
>
> Question: Could this be a 32bit vs. 64bit thing (I
> did not write any architecture specific C code, standard
> ANSI-C with double type arrays and matrices)? Is this
> a difference between how memory is handled between
> the two ports? I know that there could be a difference
> in the way the R program is written for the two
> architectures, but aside from that, is there something
> at the OS level that could be one explanation for this?
>
> Again, sorry to go off-topic, but I enjoy reading and
> respect the opinions of the contributors to this list.
>
> Many thanks,
> Chris
>
> P.S. email me personally if this question is too off
> the deep end for the list.
>
