Re: PFE Simulation Results

Small update, I've re-run the simulations using the latest version of
harfbuzz (2.6.7). This supports more lookup types and thus is a more
accurate representation of performance of patch subset. I've updated the
above doc with the updated results. As expected the total bytes transferred
increased for Arabic, Indic, Latin, Greek, Cyrillic, and Thai simulations.
This also resulted in some increases in the costs. However, the overall
results have not changed significantly.

On Wed, Jun 3, 2020 at 10:12 AM Garret Rieger <grieger@google.com> wrote:

> As discussed in our last meeting I have completed a run of the simulation
> on real page walk sequences across a variety of languages. The results of
> the simulation can be found here:
>
>
> https://docs.google.com/document/d/1aShjR0-UVoQcu1R9GGPF_Qa3bViOdqYVlk7HM07IdGU/edit?usp=sharing
>
> I'm also planning on releasing the data set that was used to drive the
> simulation. I'm hoping to have that published later this week.
>