Re: a big limit of mathematica?

On Tue, 20 Sep 2005 09:31:47 -0700, Richard J. Fateman wrote:
> Mathematica's limitations are very likely related
> to the amount of virtual memory it is allocated by
> the operating system, and perhaps by the maximum
> amount of memory Mathematica declares it is going to
> use. Such limits are much smaller than the 64 bit
> limit on address space. In particular,
> doing

That's not the way virtual memory works, at least on any decent operating
system. The amount of virtual memory that is allocated to an application
by the operating system is determined dynamically and is limited only by
the total amount of swap space available on the computer.

> Table[0,{i,1,10^4},{j,1,10^4}] 4 times on a 32-bit system
> probably gives the SAME error. But if LumisROB were to quote
> the FULL ERROR MESSAGE, it might help. The message suggests
> that quitting other applications might help.
> Maybe yes, maybe no, but full disclosure suggests this
> information should be passed on. It is still not what I'd like..
> but see below.

I just ran Table[0.,{i,10^4},{j,10^4}] on two 64-bit systems (an IBM SP
and a Mac G5) and both of them ran successfully to completion. The value
of MemoryInUse turned out to be 802872240 on the SP and 803660664 on the
Mac after the array was built. This is around 800 MB, not 4 GB.

Notice that 0. (with a decimal point, and hence a machine-size real
number) takes more space than 0 (without a decimal point, hence an
integer).

I tried it again with Table[0.,{i,10^4},{j,10^5}]. This time the SP gave
up almost immediately, but the G5 finished the computation and reported
MemoryInUse[] to be 9203663264. To be fair, this is comparing
Mathematica 5.1 on the SP with Mathematica 5.2 on the G5. However,
Mathematica 5.1 on the SP is a 64-bit app (it refuses to run on 32-bit
AIX hosts), and the SP node where I ran the test has 64 GB of physical
memory, compared to only 8 GB for the G5. That means the G5 successfully
computed an array that exceeds the size of its physical memory.


--
Dave Seaman
Judge Yohn's mistakes revealed in Mumia Abu-Jamal ruling.
<http://www.commoncouragepress.com/index.cfm?action=book&bookid=228>
.

Relevant Pages

  • Re: [Lit.] Buffer overruns
    ... > floating point support or a memory expansion option. ... had virtual memory support grafted on. ... > where the modified instruction was fetched from. ... vis-a-vis the official coporate strategic operating system TSS/360. ...
    (sci.crypt)
  • Re: Physical to virtual address mapping
    ... and memory management units and how the mapping between program ... on your processor works e.g. by studying how your operating system ... virtual memory functionality. ...
    (comp.realtime)
  • Re: MSAS 2gb/3gb limit...
    ... You said that the memory an application can ... 2-3gb Virtual Memory. ... it also relates to recommended physical memory used by ... > This is because Windows as an operating system doesn't really like to page. ...
    (microsoft.public.sqlserver.olap)
  • Re: Swap space
    ... > twice the amount of RAM if you need to capture a dump for debugging. ... > If you won't ever be doing that, you may not need so much swap. ... least the size of physical memory. ...
    (freebsd-questions)
  • Re: A big "Thank you" to databaseben, Richard and Cliff
    ... I download the smaller files first. ... small number and I had earlier bought a large amount of additional memory, ... I was able to "boost" up the virtual memory. ...
    (microsoft.public.windowsxp.help_and_support)