On Thu, Oct 1, 2009 at 10:11 AM, Erik Kay <[email protected]> wrote:

>
> On Thu, Oct 1, 2009 at 8:49 AM, Mike Belshe <[email protected]> wrote:
> > see about:tcmalloc (credit to sgk) - which dumps the *browser* stats.  We
> > need to plumb other processes.
>
> This is really cool.  My question of cost to compute was whether it
> makes sense to get this into end-user histograms, or if we can measure
> this periodically over a page cycler run.
>

Ideally yes.  I think the best way forward at this point is to define a set
of clear, measurable criteria and then try out different allocator schemes
to see what makes sense.  I'd propose the following as a starting point:

1.) Minimal performance impact
2.) Amount of committed memory never exceeds memory used by the application
by more than X% for more than Y seconds
3.) Peak memory usage on benchmark never exceeds Z MB

We don't have code currently to measure 2 and 3 exactly right now, so that's
the first priority.  After that it's a matter of picking X, Y, and Z for the
benchmarks we care about.  Any suggestions?

- James


> As you say, getting it into the other processes seems like the higher
> priority.
>
> Erik
>
>
> > Mike
> >
> > On Thu, Oct 1, 2009 at 8:42 AM, Erik Kay <[email protected]> wrote:
> >>
> >> On Wed, Sep 30, 2009 at 11:22 AM, James Robinson <[email protected]>
> >> wrote:
> >> >
> >> > I agree completely that this seems to be an issue Here's what
> >> > about:tcmalloc
> >> > says about my browser process right now (which is at around 267MB
> >> > according
> >> > to the app's Task Manager):
> >> >
> >> > MALLOC:    207097856 (  197.5 MB) Heap size
> >> > MALLOC:     12494760 (   11.9 MB) Bytes in use by application
> >> > MALLOC:    188563456 (  179.8 MB) Bytes free in page heap
> >>
> >> This is pretty eye-popping data.
> >>
> >> How expensive is it to compute this data?  Is it something we can
> >> report back in histograms?  Is it something we can add to some of our
> >> page cycler tests as perf data so we can track improvements and
> >> regressions?
> >>
> >> Erik
> >
> >
>
> >
>

--~--~---------~--~----~------------~-------~--~----~
Chromium Developers mailing list: [email protected] 
View archives, change email options, or unsubscribe: 
    http://groups.google.com/group/chromium-dev
-~----------~----~----~----~------~----~------~--~---

Reply via email to