repeat malloc & memory fragmentation

[spasmous]

Yo,

I'm running a subroutine that mallocs a 10 or so Mb, then frees it on
exit. Then when it's called again it mallocs/frees again, and so on
for maybe 100 calls.

Sometimes the code fails after running a long time - is it possible
that malloc needs a contiguous block of memory but my frequent mallocs
have fragged it all up and so the later mallocs fail. Or... am I just
a bad programmer ;)

Thanks for your input.

[lilburne]

spasmous wrote:

Sometimes the code fails after running a long time - is it possible
that malloc needs a contiguous block of memory but my frequent mallocs
have fragged it all up and so the later mallocs fail.

Yes malloc will require a contigous block of memory.

[David Harmon]

On Sun, 01 Feb 2004 19:41:06 +0000 in comp.lang.c++, lilburne
<li******@godzilla.net> was alleged to have written:

spasmous wrote:

Sometimes the code fails after running a long time - is it possible
that malloc needs a contiguous block of memory but my frequent mallocs
have fragged it all up and so the later mallocs fail.

Yes malloc will require a contigous block of memory.

But it will more often be found that the program has a memory leak
(forgetting to free something that should be) and is actually running
out of available.

[lilburne]

David Harmon wrote:

On Sun, 01 Feb 2004 19:41:06 +0000 in comp.lang.c++, lilburne
<li******@godzilla.net> was alleged to have written:

spasmous wrote:

Sometimes the code fails after running a long time - is it possible
that malloc needs a contiguous block of memory but my frequent mallocs
have fragged it all up and so the later mallocs fail.

Yes malloc will require a contigous block of memory.

But it will more often be found that the program has a memory leak
(forgetting to free something that should be) and is actually running
out of available.

Maybe, but then he doesn't say that the program has started
paging prior to running out of memory.

[David Harmon]

On Sun, 01 Feb 2004 20:35:38 +0000 in comp.lang.c++, lilburne
<li******@godzilla.net> was alleged to have written:

But it will more often be found that the program has a memory leak
(forgetting to free something that should be) and is actually running
out of available.

Maybe, but then he doesn't say that the program has started
paging prior to running out of memory.

Maybe, but he doesn't say much. Is there any more reason to think the
program would resort to paging to satisfy memory requests when the heap
was full than it would when the heap is fragmented?

[Paul]

spasmous wrote:

Yo,

I'm running a subroutine that mallocs a 10 or so Mb, then frees it on
exit. Then when it's called again it mallocs/frees again, and so on
for maybe 100 calls.

Sometimes the code fails after running a long time - is it possible
that malloc needs a contiguous block of memory but my frequent mallocs
have fragged it all up and so the later mallocs fail. Or... am I just
a bad programmer ;)

Thanks for your input.

How about looking at what you're doing between the calls of malloc and
free? Maybe something in there is causing the problem. To convince
yourself, why not write a very simple program that just allocates and
frees 10 Meg of memory in a loop?

#include <cstdlib>

void allocate_stuff()
{
char *p = (char *)malloc( 10000000 );
free(p);
}

int main()
{
char *p;
for ( int i = 0; i < 100; ++i )
allocate_stuff();
}

This code does what you describe. Does it cause problems as you've
described them, or does it run smoothly?

Paul