[minutes] 20110302 Web Performance WG

Web Performance Working Group
02 Mar 2011

See also: IRC log<http://www.w3.org/2011/03/02-webperf-irc>

Minutes
http://www.w3.org/2011/03/02-webperf-minutes.html
Attendees
Present
[Microsoft], +1.650.253.aaaa, +1.650.450.aabb, +1.650.214.aacc, +1.650.214.aadd, +1.650.704.aaee, AndersonQuach, ArvindJain, Zhiheng, KarenAnderson, JasonWeber, TonyG, JamesSimonsen, JatinderMann
Regrets
Chair
ArvindJain, JasonWeber
Scribe
AndersonQuach
Contents

  *   Topics<http://www.w3.org/2011/03/02-webperf-minutes.html#agenda>
     *   Review the proposed changes to the charter<http://www.w3.org/2011/03/02-webperf-minutes.html#item01>
     *   Navigation timing<http://www.w3.org/2011/03/02-webperf-minutes.html#item02>
     *   User Timing<http://www.w3.org/2011/03/02-webperf-minutes.html#item03>
     *   Resource Timing<http://www.w3.org/2011/03/02-webperf-minutes.html#item04>
     *   Summary<http://www.w3.org/2011/03/02-webperf-minutes.html#item05>

                 i.            Anderson will update the test_navigation_timing_order tests according to the feedback.

               ii.            Anderson will update the User Timing spec to include the updated measures and marks as discussed: above the fold, fully visible and time to user action.

              iii.            Zhiheng will update the resource timing to reflect the latest discussions on the Event Log.

             iv.            All, to think about how to bring cohesion to the Navigation, Resource and User timing interfaces.

               v.            All, to think about how to enable 'session' timings, that is a collection of individual navigation timings, whether it is with the use of existing platform functionality like DOM Storage or indexed db.

________________________________
Review the proposed changes to the charter

http://www.w3.org/2011/02/webperf.html

AndersonQuach: Are there additional deliverables?

ArvindJain: No, this has been set to review. I will send an email to request for additional items. I don't have any additional things myself.

AndersonQuach: Do we have editors?

JasonWeber: I know that Microsoft is willing to help be co-editor with each of the specs.
Navigation timing

AndersonQuach: No additional tests for this week, Microsoft is in the process of being updated.
... If there is no work around are we okay with a manual back_forward?

TonyG: If there is no workaround I'm okay with a manual test.
User Timing

move to agenda 2

rrsagenda move to agenda 2

rrsagent move to agenda 2

http://dvcs.w3.org/hg/webperf/raw-file/tip/specs/UserTiming/Overview.html

AndersonQuach: to be added, an updated return type to getMarks and the addition of the measures interface
... What's the guidance for Fully Visible and Fully Interactive?

TonyG: One pattern I see is that UI is visible but handlers are not hooked up.
... For example, Google calendar and Google search they stub out that behavior.

Zhiheng: We have a couple of things like above the fold and fully loaded. Those are the potential marks we can add. What kind of best practice can we follow-up with this draft.

TonyG: The way i'm looking at these, is an analytic script would be to look at what's in marks and measures to look specifically for items to report back. Any other mark or measure people will have to treat as a black box.

KarenAnderson: Is there value in having measures to be DOM specific?

Zhiheng: I propose two additional standard marks, Above the fold time Google services attempt to measure this. Time to User action, how long after a page load that a user will interact with the page.

AndersonQuach: Zhiheng can you provide a brief 1-2 sentence on these two standard marks, Above the fold and Time to User action.

TonyG: Should user action be in navigation timing?

JasonWeber: There are two intepretations, first when the page can be interacted with and second when the user actually interacted with the page.

TonyG: What if I want to do a measure, and navigationStart and navigationStart is zeroed out?

AndersonQuach: I'm okay with zero or an exception thrown.

JasonWeber: For common practice, zero may be better as exceptions may lead to more complicated code.

TonyG: I'm okay with zeros.

JasonWeber: Perhaps we can return undefined.

Zhiheng: I agree, zero might be some valid value in some cases.

TonyG: When you call measure, it will create a mark and measure. Mark there is no problem.

AndersonQuach: I will capture all these caveats into the spec.
Resource Timing

move to agenda 3

AndersonQuach: Tony and James have you gotten feedback with Resource Timing?

JamesSimonsen: No.

TonyG: We're amicable about the event log. However we should consider off by default.

JasonWeber: I think that's a valid concern. Do you have a projection of what those resource costs would be in Chrome.

TonyG: In terms of raw data, it's the same in IE. It depends on the estimates, on the resources on a page.

JasonWeber: There's the working set resources which we estimated that impact. There's the alloc patterns we don't believe have an impact to performance. There's the CPU and contention, we already have the data it's easy for us to store that data. We don't expect contention and low CPU usage.
... If this is an arch concern for Webkit and Firefox and influence the off by default and on by default conversation. We look at it today and don't see the resource concern. If you have projections about slowing down Webkit and Chrome that can help us steer this discussion.

JamesSimonsen: Arch from the Chrome side, I don't see an issue. On a mobile browser, these become a concern.

JasonWeber: IE is in the same position. IE on mobile is excited to get access to this data.

TonyG: One thought, agree on the event log approach, can be enabled or disabled by default. It's recommended enabled by default. We have methods in the interface start-stop tracking of resources. Implement the interface so it's there. Experiment with enable by default.

JasonWeber: Ya.
... That sounds like the right approach.

Zhiheng: That makes sense too.
... How can we enable / disable, the web-site opt-in.

TonyG: Ya, the website opts in, but does not allow for retro-active analysis. JavaScript method, http header or meta tag.

KarenAnderson: Is there concern of disclosure?

TonyG: For sure there are privacy concerns. We can't expose timings for resources on other domains. We have to work through.

AndersonQuach: Jason Sobel has been requesting for a blessed way of access cross-domain timings.

TonyG: Do we want to have Resource Timing and User Timing in the same structure?

AndersonQuach: Can you clarify?

TonyG: Was wondering if we can tie all the logs together.
... Is there a way to efficiently bring back all the data when a web-developer wants navigation, resource and user timing altogether.

AndersonQuach: Two thoughts, Resources timing can be filtered by type and User Timing may be larger than navigation timing to separate the data serializing.

JamesSimonsen: Can user timing and resource timing be different types?

AndersonQuach: We could have getTimings versus getMarks

Zhiheng: User Timing, would love to have a way to pass user timing across navigations.
... It's a minor value add, especially being able to time across navigations.

AndersonQuach: Are the scenarios like log-on or cross page scenarios.

Zhiheng: That's already achievable by using storage solutions.
... Bring this up to get some feedback.

AndersonQuach: This is something we're actively mitigating with cross-domain accesses.
Summary

i. Anderson will update the test_navigation_timing_order tests according to the feedback.

ii. Anderson will update the User Timing spec to include the updated measures and marks as discussed: above the fold, fully visible and time to user action.

iii. Zhiheng will update the resource timing to reflect the latest discussions on the Event Log.

iv. All, to think about how to bring cohesion to the Navigation, Resource and User timing interfaces.

v. All, to think about how to enable 'session' timings, that is a collection of individual navigation timings, whether it is with the use of existing platform functionality like DOM Storage or indexed db.