<chris> scribenick: ned

scribenick ned

vlad: housekeeping

chris: what to do about asymmetric connection speeds

<chris> assymetric speeds can wait I think

<Vlad> close action-221

<trackbot> Closed action-221.

garret: [reviewing report]
... report now includes range requests
... new data set, much bigger than previous
... samples for languages with too few data points before
... same data source as before but wider sampling
... new graphs showing language breakdown of dataset
... LGC+Thai
... Arabic, also good selection of Indic scripts
... CJK was previously almost completely Japanese
... now have more diversity in CJK set
... [graphs]
... box plots now show 25/50/75 percentiles
... patch subset generally greater bytes reduction, patch subset better on 2G

[sorry, range request better on 2G]

garret: Arabic/Indic patch subset similar to Unicode range, patch subset has greater reduction of bytes transferred
... LGC+Thai patch subset worse than Unicode range
... conclusion is that patch subset is an improvement over Unicode range

myles: "pie" graph is maybe not relevant to interpretation of results

garret: groups being compared are LGC+Thai, Arabic+Indic, and CJK

myles: suggests putting pie charts with box plots for each language group

[discussion of this report's relation to evaluation report]

myles: rather than providing feedback on this document, maybe I should provide feedback on evaluation report
... what about cost function? need better understanding

chris: +myles

myles: current cost function is arbitrary

<chris> cost function is described here https://docs.google.com/document/d/1kx62tpy5hGIbHh6tHMAryon9Sgye--W_IsHTeCMlmEo/edit#heading=h.4fz1x8661i63

garret: analysis framework describes current (second) cost function

myles: bytes might be more valuable than percentage

garret: include average font sizes?

myles: yes
... as far as colors go, would like more explanation
... hard to distinguish between all results being good and some being better than others
... don't forget about non-performance results, like Unicode range breaking layout
... conclusion is slightly unfair since it doesn't address non-performance criteria

garret: conclusion is only meant to address performance and not final recommendation

myles: I have feedback written down, will make sure nothing is lost

<myles> 1. distribute language breakdown pie charts into the language group sections

vlad: next item

<myles> 2. Needs more discussion about the arbitrary cost function. Need to discuss the difference between cost vs bytes, and how users should consider them when reading the graphs

<myles> 3. Need to include information about relative font sizes and bytes saved between different language groups. The big fonts are the ones that are important to optimize, that information is lost here

vlad: how many connections are symmetric?

garret: except for cellular, most are asymmetric

<myles> 4. It was non obvious to me to determine how to read the graphs. Needs to be described better. Green is good, red is bad. Graphs that show all green show that all methods provide improvement.

garret: didn't want to use inaccurate data and so assumed up/down to be equal

<myles> 5. Evaluation report needs to include non-performance information too (it's super important). E.g. unicode-range method breaks shaping, patch subset needs to run software on a server

chris: ADSL example

vlad: cable also

chris: ADSL more pronounced

vlad: says you

<myles> 6. Similar to point 5: The conclusion shouldn't recommend any option because none of this non-performance info is considered

chris: should use established broadband specs as realworld examples of up/down speeds

myles: my tests also assumed equal speeds

<myles> 7: We need a better description of why the 2G graphs are separate. The reason why is a good reason, it just needs to be stated

garret: not sure it makes much difference, but Unicode range requres many more requests so either new method is going to be roughly equivalent

vlad: big meetings coming up in October, overdue to publish result
... if results based on educated guesses are sufficient, redoing tests may be overkill

chris: can note assumption in report

<chris> https://w3c.github.io/PFE-analysis/report/evaluation-report.html

chris: need to say a lot about fonts!

myles: TrueType and OpenType are different things

<myles> ned: We have the distinction between SFNT, and the data inside it. There's a lot of overlap between OT & TT. In OT, CFF is a completely self-contained font, but in TT, the glyf table is tied to hmtx table.

<myles> chris: CFF2 is not a self-contained font, right?

<myles> ned: Yes, it reduces the redundancy. If we're talking about transfering parts of a font, we need to mention that there is that difference in terms of the dependencies.

<myles> chris: Range request is harder for CFF because there is duplicated information in CFF and the rest of the font

<myles> chris: Is that information the same?

<myles> ned: it's "same enough"

<myles> chris: are the graphs helpful?

chris: may flip colors in map

<myles> myles: yes. I like the map

<myles> chris: I will flip red and green

<myles> chris: red vs green, i mean

<myles> <conversation turns to discussing colorblindness>

<also Hangul>

<myles> jpamental: Would be good to have a graph showing fonts' file sizes and glyph counts for the various language groups

<myles> chris: yes

myles: running out of time

vlad: perfect

This is scribe.perl Revision of Date 
Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/

Guessing input format: Irssi_ISO8601_Log_Text_Format (score 1.00)

Succeeded: s/file sizes and glyph counts/file sizes and glyph counts for the various language groups/
Present: jpamental Vlad sergeym Garret myles chris

WARNING: Replacing previous Regrets list. (Old list: Persa, (birthday))
Use 'Regrets+ ... ' if you meant to add people without replacing the list,
such as: <dbooth> Regrets+ Persa, Zula

Regrets: Persa Zula
Found ScribeNick: ned
Inferring Scribes: ned

WARNING: No "Topic:" lines found.


WARNING: No meeting chair found!
You should specify the meeting chair like this:
<dbooth> Chair: dbooth

Found Date: 28 Sep 2020
People with action items: 

WARNING: Input appears to use implicit continuation lines.
You may need the "-implicitContinuations" option.


WARNING: No "Topic: ..." lines found!  
Resulting HTML may have an empty (invalid) <ol>...</ol>.

Explanation: "Topic: ..." lines are used to indicate the start of 
new discussion topics or agenda items, such as:
<dbooth> Topic: Review of Amy's report


WARNING: IRC log location not specified!  (You can ignore this 
warning if you do not want the generated minutes to contain 
a link to the original IRC log.)