| Re: Clear everything in IDL? [message #9932] |
Wed, 24 September 1997 00:00  |
Peter Mason
Messages: 145
Registered: June 1996
|
Senior Member |
|
|
On Tue, 23 Sep 1997, Alex Schuster wrote:
> Matthew J. Sheats wrote:
>> I've tried this.. unfortunately ( In windows NT anyway.. I haven't
>> tried else where ) It does not seem to actually release the memory as
>> such.
>> Internal to IDL, it may mark that region of memory as freed, but
>> Windows never receives that memory back until IDL is closed. It would
>> also appear that any additional memory allocated doesn't utilize this
>> memory which idl has already allocated, but instead pulls more off of
>> NT's heap.
>
> You're right! The IDL FAQ has a topic covering this. Have a look at
> http://ww2.sd.cybernex.net/~mgs/idl_faq.html#T27
This is a problem that Unix IDL users know and hate - the IDL session's "data
space" never gets reduced in size. (It's an operating-system issue, not
an IDL issue.)
Now try not to laugh, but if this is a major issue for you then you might be
better off using Windows 95 instead of NT. Win95 appears to release memory
back to the system as soon as it is "freed" in IDL. (I must say that I'm
disappointed to hear that NT seems to have the same dog-with-a-bone problem as
Unix.)
If you are swinging huge arrays around in IDL then it'll probably be worth your
while to explore the way IDL uses memory "on the sly". This is easy under
Win95. Start the System Monitor and get a "Memory Manager: Allocated Memory"
display going, and start an IDL session. Get yourself a chunky array, e.g.,
N=1024L*1024L*32L & B=BYTARR(N). Try out some stuff. e.g., Watch your
computer go to lunch when you type B(*)=1B. Enjoy the German engineering of
B=TEMPORARY(B)+1B. Look on in horror at B=TEMPORARY(B)+1. See if your
computer is even capable of B(*)=1L. And try out some of the operations your
program does on large arrays.
Peter Mason
CSIRO division of Exploration and Mining
P.O Box 136, North Ryde, NSW, 2113, Australia
E-Mail: p.mason@syd.dem.csiro.au Tel: +61 2 9490-8883 Fax: 9490-8960/8921
Web: http://www.syd.dem.csiro.au/research/MMTG/
|
|
|
|
|
|
|
|
|
|
| Re: Clear everything in IDL? [message #10010 is a reply to message #9949] |
Fri, 26 September 1997 00:00  |
Peter Mason
Messages: 145
Registered: June 1996
|
Senior Member |
|
|
On Wed, 24 Sep 1997, Matthew J. Sheats wrote:
> There is a bigger issue here unfortunately. At this time I'm not so
> sure if it is an OS related problem. I can write programs dynamically
> allocating memory from the heap and releasing it back without a hitch in
> C++. So I think there is more going on. Perhaps ( as the FAQ says ) it
> is because they use malloc's etc.
Fraying the thread...
It looks like IDL 5 is using a 3rd-party memory-allocation library called
"SmartHeap" (on Win95 at least). This purports to be far quicker and more
efficient than the standard malloc library you get with a compiler.
Well, apparently not always. I'd guess that it must be a help for the
countless reasonably-small allocations a typical IDL program does, but it
appears to get in the way when really chunky allocations (in the megs) are
involved.
With IDL5/Win95 I still see "freed" memory (*large* variables) getting returned
to the system. However, really large allocations seem to take a lot longer
than in IDL4 (which doesn't use SmartHeap, I presume), and the ceiling for
allocations appears to be significantly lower. (Well, no, it *is* lower - I
can definitely get away with more in IDL4.)
Peter Mason
(Mr. Subjunctive)
|
|
|
|
| Re: Clear everything in IDL? [message #10024 is a reply to message #9949] |
Wed, 24 September 1997 00:00 |
Matthew J. Sheats
Messages: 19
Registered: September 1997
|
Junior Member |
|
|
Peter Mason wrote:
>
> This is a problem that Unix IDL users know and hate - the IDL session's "data
> space" never gets reduced in size. (It's an operating-system issue, not
> an IDL issue.)
> Now try not to laugh, but if this is a major issue for you then you might be
> better off using Windows 95 instead of NT. Win95 appears to release memory
> back to the system as soon as it is "freed" in IDL. (I must say that I'm
> disappointed to hear that NT seems to have the same dog-with-a-bone problem as
> Unix.)
>
There is a bigger issue here unfortunately. At this time I'm not so
sure if it is an OS related problem. I can write programs dynamically
allocating memory from the heap and releasing it back without a hitch in
C++. So I think there is more going on. Perhaps ( as the FAQ says ) it
is because they use malloc's etc.
If the program I am writing was just for my personal use, I would just
run on 95 and be happy. Unfortunately, I'm building it to be used on
any architecture that IDL supports. It would be nice to have
predictable behavior across architectures and OS's.
So I guess I'll write for the lowest common denominator until some kind
of resolution can be found.
And thanks to everyone participating in this discussion, it's been
interesting.
Matthew J. Sheats
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
