Re: Graphs for Final Font Size

File size distribution graph is here:
https://github.com/w3c/PFE-analysis/pull/57. This only covers the full
woff2 file sizes. It's pretty hard to see the lgc and arabic ranges, but I
think it pretty effectively communicates the huge difference in CJK sizes
vs non-CJK fonts.

On Mon, Oct 12, 2020 at 2:34 PM Garret Rieger <grieger@google.com> wrote:

> It may be a good idea to point out the two caveats mentioned above when
> presenting the bytes reduction graphs in the evaluation report to avoid
> misinterpreting those numbers as the reductions expected for a single font
> viewed in isolation.
>
> On Mon, Oct 12, 2020 at 2:32 PM Garret Rieger <grieger@google.com> wrote:
>
>> Here's my attempt at creating a graph that has the total bytes
>> transferred for a font across the three tested methods:
>>
>> https://github.com/w3c/PFE-analysis/pull/56/files
>>
>> Some notes:
>>
>>    - I stuck with the same style graph as used in the percentage graphs
>>    as it lets us easily show distributions. I changed the line colour to blue
>>    since there's no good/bad distinction on this graph.
>>    - I derived the values on the graph by applying the percentage
>>    reductions from the simulation results to the "worst case" woff2 font size
>>    for each language group (the 95th percentile font size).
>>    - This gives an estimate of the amount of bytes that would be
>>    transferred for a font of that size.
>>    - However, there are a couple important caveats:
>>       - % reduction from the simulation is the reduction over a sequence
>>       of page views and is *not representative of the reduction in bytes
>>       transferred for loading a single page*. The reduction for a single
>>       page is likely to be higher than for a sequence of page views.
>>       - Similarly the reduction is the aggregate of all fonts on a given
>>       page. So for something like CJK that means there could be latin or other
>>       language fonts mixed in. Thus the percent reductions *are not
>>       representative of the reductions you'd see for a single CJK font in
>>       isolation*. Again I believe the reduction is actually larger for a
>>       single CJK font on a single page.
>>    - Unfortunately the simulation results were not recorded at the font
>>    level but at the page view level, so there isn't a way to the true per font
>>    reduction percentages out of the results we currently have.
>>
>>
>> Next up I'm working on generating graphs of the font size distributions.
>>
>>
>