Chris Waterson wrote:
>
> Updated [1] to include allocator performance comparison between glibc
> and Lea-2.7.0pre6. (Still need to collect data for pre7, and have to
> respin data for BSD and Hoard because some of the growth problems have
> purportedly been fixed in the latest releases.) Anyway, bottom line is
> that gtkEmbed page load performance using lea-2.7.0pre6 is ~3% slower
> than with the glibc allocator.
>...
>From the there-are-three-kinds-of-lies dept.: You can't be confident
about that difference from the results you posted the graph of (or the
updated graph on <http://mozilla.org/projects/footprint/allocators.html>
either) -- because the value for glibc is within the confidence interval
of lea for many of the URLs in the graph.
If you combined the results of all the URLs together, the confidence
interval would probably shrink enough so that you could be sure of the
advantage of using glibc.
Or, you could just decide to be less confident (e.g. 90 percent rather
than 95 percent).
--
Matthew `mpt' Thomas, Mozilla user interface QA
Mozilla UI decisions made within 48 hours, or the next one is free
