Greg,
Just out of curiosity how much memory should it take to import a ~50MB
tree? It seemed "reasonable", not optimal, but reasonable to me to
consume 10-12MB. It appears that there are a number of pools and
sub-pools which are only cleared on a per directory basis from what I
could tell so the bigger and deeper the tree the more memory it
consumes. I haven't tested Sander's patch yet to see how much is used
in a production environment. It appears the the pooling code just
wasn't giving up the memory you told it could, assuming it could reuse
it from Sander's description.
Anyways, I will try that import/check in as many versions of the Linux
kernel as I have space for and then check them back out and see what the
memory consumption is on import/ci/co with the production w/ patch
versus debugging code and see what falls out.
Thanks,
Kirby
Greg Stein wrote:
> On Sun, Feb 03, 2002 at 02:17:02AM +0100, Sander Striker wrote:
>
>>Given the reports about high memory usage, a small
>>patch (bandaid). I'll be looking into a more permanent
>>solution tomorrow and upcomming week.
>>
>
> I believe this only patches the symptom, rather than the underlying problem.
> It seems that we shouldn't be using all that much memory in the first place
> -- only enough to persistently hold some metadata. Anything outside of
> metadata should be handled streamily. And some of that metadata should be
> recycled.
>
> Cheers,
> -g
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Sat Oct 21 14:37:03 2006