RE: documentation for boehm-gc tuning/profiling? [more info]

["Boehm, Hans" <hans_boehm at hp dot com>]

Just setting the GC_PRINT_STATS environment variable may give you some
information.  Or it may run into the same problem as without -DSILENT,
whatever that is.

What does "make check" in the boehm-gc directory do?  Does gctest run at
reasonable speed (a few seconds)?

Can you get gdb to stop the program while it's running slowly to at least
get some stack samples?  If you can somehow call GC_dump() that might also
help.

It sounds like some memory region is getting far too big.  It may be that
the collector is OK but the size of allocation requests (i.e. the size
argument to GC_gcj_malloc) is somehow getting garbled.  I would be really
nice to look at the stack somehow.

Hans


-----Original Message-----
From: Adam Megacz
To: java@gcc.gnu.org
Sent: 12/22/01 5:58 PM
Subject: Re: documentation for boehm-gc tuning/profiling? [more info]


Adam Megacz <gcj@lists.megacz.com> writes:
> Jeff Sturm <jsturm@one-point.com> writes:
> > On 22 Dec 2001, Adam Megacz wrote:
> > > So things are going pretty well with my win32 port of xwt, except
that
> > > allocating new objects seems to be incredibly time-consuming -- up
to
> > > (I'm guessing) around 200ms just to concatenate two strings. It
seems
> > > to take longer the larger the heap is.

Also, when I cause my app to engage in very minimal memory allocation
(say, five instantiations per second), sysmon.exe reports page faults
jumping from 0 pagefaults/sec (at rest) to 64 pagefaults/sec.

The number of locked non-cache pages also increases a bit.

 - a