thanks much for your response, off hand do you know
any process analysis tools that would give me this
information?

Tom



--- Graeme Roy <graeme@...> wrote:

> Hi Tom,
>
> The short answer is that it doesn't free the memory
> back to the system, and
> neither do most other malloc libraries.
>
> And now for the long answer...
>
> For systems using sbrk() (which mpatrol uses on UNIX
> by default), it would
> be virtually impossible to return the memory back to
> the system due to the
> fragmentation of the heap. The only way you could
> really do this would be
> to get the malloc library to free each of the pages
> it allocated in reverse
> order by calling sbrk() with a negative value, but
> that would also require
> the user's program to do frees() in the same way.
> I'm not saying it's
> impossible to do because I'm sure some programs
> would lend themselves to
> that, but in most cases it would be impossible and
> malloc libraries have to
> handle the general case.
>
> For memory allocated by mmap() (which mpatrol uses
> with USEMMAP) or
> VirtualAlloc() on Windows, the situation is slightly
> less complicated
> because the allocated pages could in theory be freed
> in any order if there
> are no more malloced allocations located in them.
>
> However, there are two main reasons why malloc
> libraries won't free the
> memory back to the system.
>
> The first is that system calls are expensive
> operations and who's to say
> that once you've instructed the system to free up a
> chunk of memory from the
> process, you won't have to then claim that memory
> back from the system next
> time the process calls malloc(). It's much more
> efficient just to maintain
> that claimed memory in the free pool in case the
> process needs it later.
>
> The second is that if you're running your process on
> an operating system
> that supports virtual memory (and I suspect most
> people are, except those on
> most embedded systems) it's really virtual memory
> rather than physical
> memory that's been allocated. Which means that if
> the page isn't used for a
> while (i.e. it doesn't contain any memory
> allocations in use by the process)
> then it will be moved out of the working set of the
> process and possibly out
> of physical memory until it's next needed. There's
> some more information on
> virtual memory at:
>
>
http://www.cbmamiga.demon.co.uk/mpatrol/mpatrol_13.html
>
> I did originally think about adding a feature to
> mpatrol to free the pages
> back to the system when they weren't needed, but the
> extra work involved put
> me off, not to mention the slowdown that would
> result, and in the end I
> decided it was probably best to mimic what existing
> malloc libraries do.
> Plus, it meant that I could go ahead and implement
> the NOFREE option, which
> I think was a good tradeoff.
>
> To answer your other question, mpatrol only asks the
> system for more memory
> when it can't find a suitable-sized free allocation.
> It uses sbrk() on UNIX
> by default but with USEMMAP it will use mmap(). On
> Windows it always uses
> VirtualAlloc(). You can read a bit more about how
> mpatrol works at:
>
>
http://www.cbmamiga.demon.co.uk/mpatrol/mpatrol_42.html
>
> mpatrol will only really tell you about the
> high-level memory allocations
> with malloc(), operator new, etc. If you're wanting
> to profile the heap of
> a process on a low-level page by page basis then
> you're probably better off
> using a process analysis tool, especially if you're
> relying on using a
> specific malloc library.
>
> Hopefully that's answered your queries.
>
> Thanks,
>
> Graeme.
>
>
> -----Original Message-----
> From: mpatrol@yahoogroups.com
> [mailto:mpatrol@yahoogroups.com] On Behalf Of
> tgraves_cs
> Sent: 22 November 2006 16:27
> To: mpatrol@yahoogroups.com
> Subject: [mpatrol] mpatrol heap efficiency ?
>
> Hello all,
>
> I'm new to using mpatrol and I'm trying to use it to
> figure out heap
> efficiency. Basically when the process mallocs
> memory, reserves it,
> and then when it frees the memory and when it can
> actually free
> blocks/pages back to be used by everyone else.
>
> I see in the mptrace tool it tells you when it
> reserves it:
> 12 free 20 0x093E71AC 16 1
> 0 0
> 13 alloc 21 0x093E71AC 16
> 1 16
> reserve 0x093E9000 98304
> 14 alloc 22 0x093E7904 96000
> 2 96016
> reserve 0x09401000 98304
>
>
> But I don't see it listing when it frees it back to
> general use by
> everyone.
>
> Does anyone know if mpatrol does this? How does it
> figure out when it
> reserves it? Does it wrap sbrk?
>
> Thanks,
> Tom
>
>
>
>
> To Post a message, send it to: mpatrol@egroups.com
>
> To Unsubscribe, send a blank message to:
> mpatrol-unsubscribe@egroups.com
> Yahoo! Groups Links
>
>
>
>
>




________________________________________________________________________________\
____
Do you Yahoo!?
Everyone is raving about the all-new Yahoo! Mail beta.
http://new.mail.yahoo.com