Re: WOFF2 Glyf Transform for Arabic, Indic, and LGC simulations

One thing that complicates matters is that we use different amounts of
codepoint prediction on the different networks. Which will change how a
byte reduction affects the cost. That said I'm not sure that's the only
explanation for what we're seeing here.

On Tue, Dec 8, 2020 at 2:20 PM Levantovsky, Vladimir <
Vladimir.Levantovsky@monotype.com> wrote:

> Thank you Garret, these results are really interesting!
>
> I find the total bytes reduction of 10-15% across different scripts to not
> be unusual [and even expected], but what I found surprising is that cost
> reduction graphs do not show what I would consider an _*expected*_
> outcome: I expected that larger latency networks would essentially diminish
> the effects of bytes reduction as part of the overall costs, and that cost
> reductions would improve with faster networks. But the results show that
> both ends of the “network speed spectrum” show modest gains while 4G
> performance gains are most pronounced – I am curious why this might be the
> case but can’t seem to find a reasonable explanation. Any insights?
>
>
>
> Thank you,
>
> Vlad
>
>
>
> *From:* Garret Rieger <grieger@google.com>
> *Sent:* Monday, December 7, 2020 4:53 PM
> *To:* w3c-webfonts-wg (public-webfonts-wg@w3.org) <
> public-webfonts-wg@w3.org>
> *Subject:* WOFF2 Glyf Transform for Arabic, Indic, and LGC simulations
>
>
>
> Previously I reported the effect on performance of adding woff2
> glyf transformation to patch subset patch computation for CJK. I just
> finished running the same test against the Arabic, Indic, and LGC data
> sets. Results are looking very good for those scripts as well. We see a
> pretty similar reduction in total bytes transferred of 10-15%:
>
>
> https://raw.githubusercontent.com/googlefonts/PFE-analysis/woff2_plots/results/07-08-2020/comparison.woff2.arabic_indic.sampled_500.bytes.png
> <https://protect-us.mimecast.com/s/_NStCyPD59uN35qpsZw8Ba>
>
>
> https://raw.githubusercontent.com/googlefonts/PFE-analysis/woff2_plots/results/07-08-2020/comparison.woff2.arabic_indic.sampled_500.cost.png
> <https://protect-us.mimecast.com/s/w51PCzpE57sRzoQ7HXrK31>
>
>
> https://raw.githubusercontent.com/googlefonts/PFE-analysis/woff2_plots/results/07-08-2020/comparison.woff2.latin.sampled_1000.bytes.png
> <https://protect-us.mimecast.com/s/_n2wCAD2oMh95AL8F942tN>
>
>
> https://raw.githubusercontent.com/googlefonts/PFE-analysis/woff2_plots/results/07-08-2020/comparison.woff2.latin.sampled_1000.cost.png
> <https://protect-us.mimecast.com/s/ZqBZCBB2pPcVN5pyFjv4u0>
>