[minutes] 20101201 Web Performance Working Group

Web Performance Working Group Meeting #12
01 Dec 2010

See also: IRC log<http://www.w3.org/2010/12/01-webperf-irc>

Minutes
http://www.w3.org/2010/12/01-webperf-minutes.html
Attendees
Present
[Microsoft], +1.650.253.aaaa, Zhiheng, Philippe, +1.650.704.aabb, James, AndersonQuach, Christian', ArvindJain
Regrets
JasonWeber
Chair
ArvindJain
Scribe
AndersonQuach
Contents

  *   Topics<http://www.w3.org/2010/12/01-webperf-minutes.html#agenda>
     *   Feedback and results on the name-search for window.(performance | webPerformance | pagePerformance).<http://www.w3.org/2010/12/01-webperf-minutes.html#item01>
     *   Connect phase feedback, SSL Handshake and SOCKS authentication.<http://www.w3.org/2010/12/01-webperf-minutes.html#item02>
     *   "Unload Event feedback."<http://www.w3.org/2010/12/01-webperf-minutes.html#item03>
     *   Review a test case."<http://www.w3.org/2010/12/01-webperf-minutes.html#item04>
     *   Resource timing.<http://www.w3.org/2010/12/01-webperf-minutes.html#item05>
     *   Summary<http://www.w3.org/2010/12/01-webperf-minutes.html#item06>

A.    [Decision] Closed in window.performance namespace, moving forward with the name for future extensibility due to the low risks outlined in this meeting.

B.    [Decision] handShakeStart added as a new [optional] attribute with three states: undefined for those browsers that do not have insight to the network-stack, 0 value for non-ssl connections and a time-stamp for ssl connections. connectEnd will capture the full end-to-end time, including SOCKS and SSL handshake.

C.    [Decision] unloadEventStart, unloadEventEnd to capture the start and end of the unload event handler. The unloadEventStart already captures the beginning of the document clean-up steps as outlined in http://dev.w3.org/html5/spec/history.html#unloading-documents. The goal is to provide actionable data to site developers.

D.    Test cases should include: Title, Description / objective and clear pass / fail criteria.

E.    Review offline and provide feedback for Resource Timing proposals as outlined in James' summary email.



________________________________

<scribe> scribe:AndersonQuach

list the agenda

AndersonQuach: query results returned, 0 hits with pagePerformance, however recommend performance due to low risks, alternatively webperformance also at zero hits.

Zhiheng: want to discuss a read/write attribute rather than read-only.

NicJansma: if we remove read-only, can we revert to the browser implementation.

Christian: What is the adv. for read/write?

NicJansma: Site compat, to reduce chances for sites throwing script errors?

Christian: you can still get errors with the attribute being read/write?

plh: making it read-write any application can modify it.

Zhiheng: performance makes more sense to me.
... Let's move forward with performance, the risks seem to be low.
Feedback and results on the name-search for window.(performance | webPerformance | pagePerformance).

Zhiheng: Decision to continue to use window.performance.
Connect phase feedback, SSL Handshake and SOCKS authentication.

Zhiheng: I agree for connect to include end to end time, including tcp connection, socks and ssl.

AndersonQuach: Let's not include SOCKS, does not pertain to the website.

Zhiheng: may-level requirement to have handshakeEnd == connectEnd if SSL in the picture.

NicJansma: Existence of the attribute may be confusing, if it does not work.

Zhiheng: agreed, optional for the handshake phase for user agents that no insight to the network stack.

Christian: better to be zero, if there is no handshake.

Zhiheng: there are three values: undefined, 0 for no handshake, and value when there is a handshake.
... End of handshake would be connectEnd

move to agendum 3
"Unload Event feedback."

http://dev.w3.org/html5/spec/history.html#unloading-documents

Zhiheng: I'm okay with capturing it as the unload handler as site developers have the most control over this.

move to agenda 4
Review a test case."

AndersonQuach: feedback, we should include: title of tests, description of tests, objective, and clear pass / fail criteria.
... would recommend that we pursue self-contained tests until we need a common infrastructure for testing async behavior and etc.

http://test.w3.org/html/tests/approved/canvas/canvas_colorsandstyles_addcolorstop_001.htm

AndersonQuach: this is a great start

NicJansma: Thanks for taking the lead. We'll follow offline with the recommendations for details on the tests.
Resource timing.

AndersonQuach: James, thanks for summarizing.

NicJansma: Is there a recommendation you have?

simonjam: Yes, the first proposal, attaching to the simple event.
Summary

1. Closed in window.performance namespace, moving forward with the name for future extensibility due to the low risks outlined in this meeting.

2. handShakeStart added as a new [optional] attribute with three states: undefined for those browsers that do not have insight to the network-stack, 0 value for non-ssl connections and a time-stamp for ssl connections. connectEnd will capture the full end-to-end time, including SOCKS and SSL handshake.

3. unloadEventStart, unloadEventEnd to capture the start and end of the unload event handler. The unloadEventStart already captures the beginning of the document clean-up steps as outlined in http://dev.w3.org/html5/spec/history.html#unloading-documents. The goal is to provide actionable data to site developers.

4. Test cases should include: Title, Description / objective and clear pass / fail criteria.

5. Review offline and provide feedback for Resource Timing proposals as outlined in James' summary email.