[Gc] nobody out there has experienced with USE_MUNMAP?
Hans.Boehm at hp.com
Fri May 12 07:42:18 PDT 2006
I generally haven't observed that, though a few others have commented
that it is important for them.
My reasons for not pushing it more are mostly:
- It has a potential performance down side, especially given the current
implementation. It adds system calls, and pages may be cleared
repeatedly, by the kernel, and in user space.
- I'm not particularly happy about the current implementation, which
unmaps at heap chunk instead of page granularity. This causes some
ugliness and possibly extra work when large heap blocks are merged.
If your experience is typical, I'm not opposed to making it the default.
But I suspect you are dealing with slightly unusual allocation behavior.
On Fri, 12 May 2006, Glauco Masotti wrote:
> Hello. I understand that you gurus (:-) are interested in other details of the software...
> but nobody out there has experienced the same heap growth problems that I described in my previous post?
> (Too bad that the graphs of memory management cannot be shared).
> As I said, in my case, the USE_MUNMAP option produces a much better behavior in memory management.
> What is reported in the GC documentation about this, is just what follows:
> "Heap fragmentation. This should never result in unbounded growth, but it may account for larger heaps. This is most commonly caused by allocation of large objects. On some platforms it can be reduced by building with -DUSE_MUNMAP, which will cause the collector to unmap memory corresponding to pages that have not been recently used. "
> Which is not much, and does not correspond exactly to what I observed.
> Nobody has to comment or to share experience?
> Thanks a lot in advance.
> --- Glauco Masotti
More information about the Gc