Rounding up huge allocations to page boundaries instead of chunks

[Jason Evans]

On Oct 1, 2014, at 4:07 PM, Guilherme Goncalves <ggp at mozilla.com> wrote:
> So to summarize, if I understand it correctly:
> 
> - There is ongoing work to have all allocations use the same size classes (including,
> in particular, huge allocations, which is what we're concerned with);
> 
> - The size classes will be defined as equal-sized intervals dividing each chunk boundary;
> to use your example, with 4 intervals per 4MiB chunk, we'll have [4MiB, 5MiB, 6MiB, 7MiB],
> [8MiB, 10MiB, 12MiB, 14MiB], [16MiB, 20MiB, 24MiB, 28MiB], ...;
> 
> - The number of intervals is configurable via the |lg_g| variable in size_classes.sh, which
> allows us to bound the over-reporting to a ~2^-lg_g factor of the allocated size.
> 
> If that's correct, then yes, I believe that satisfies our needs. Is there a GitHub issue I
> can follow to track progress on this? Or is there otherwise any expected timeframe for this
> to be completed?

Yes, that's all correct.  You can track progress at:

	https://github.com/jemalloc/jemalloc/issues/77

I think the hard parts are all done now, and I'm hoping to finish up this particular task within the next few weeks.  The timeframe for the 4.0.0 release is still a bit uncertain, but I'm currently shooting for the end of 2014.  You can track how that's going here:

	https://github.com/jemalloc/jemalloc/issues?q=is%3Aopen+is%3Aissue+milestone%3A4.0.0

Thanks,
Jason