Meeting 2024-10-29
-
Team calendar
- Next on Necko triage: Randell
-
New contributors / Friends of Necko THANK YOU!!
- None in the last week.
-
Round Table
- – enter public topics –
-
Active Projects roadmap Fx132 rel Oct 29, Fx133 rel Nov 26, Fx134 code freeze Nov 21
-
Firefox Essential Internal Services Toughening[FEISTY]
- [Greg] Passed internal review, now in strategic review
-
DoH Enablement - Fenix and Focus
- [Greg] Passed internal review, now in strategic review
-
3rd Party Cookie Deprecation (in progress, target Fx134)
-
3PCD timeline
-
131: Enable in nightly all windows(dry run)
- Check back once 131 hits release and see the telemetry
-
132: Enable in PBMode and ETP Strict to release
- Tracking breakage
-
133: Release telemetry on partition limit will start, review it (we are here)
- – enter updates –
-
134: All windows in release (targeting January)
-
136: “Migration”/deletion of TCP cookies (non-chips partitioned)
-
-
[Ed] Ongoing investigations and discussions with privacy team for google-auth followup
-
-
Fetch keepalive: (riding the trains, Target Fx133)
- – enter updates –
-
HTTP Priorities Experiments (riding the trains, target Fx 132)
-
[Greg] Released today closing!
-
[ACreskey] First experiment that selects only pages that don’t make use of fetchpriority (images). Promising results
-
2.6% improvement in mean lcp, up to 7% at P80
-
Following up on Fenix (nightly) and Desktop (beta)
-
[skim working on:]
-
-
Firefox/Chrome benchmark test suite (in progress, Fx132)
-
First cut of download results: https://lookerstudio.google.com/reporting/f5fbaacd-a473-4daf-bf22-faa3d3c8391f/page/DO9EE
-
Needs validation, maybe hand testing
-
Related: Chromium’s loadline framework
-
(greg) Perf Infra Hardware review Performance Hardware Review
-
– enter updates –
-
-
-
Interop 2024 URL (in progress, target H2 2024)
-
[Ed] Landed Add defaultUri bypass and Enable defaultURI
-
URL interop score moves to 99.3% of our commitments (96.8% if we consider all of URL test scores)
-
-
necko-glean: (sekim) Landing glean patches using #ifndef ANDROID blocks tracking such blocks using Bug 1901798
- – enter updates –
-
Fast UDP I/O in Firefox (in progress, target H2 2024)
-
milestone 1 - basic end-to-end integration
-
On Android x86 Focus and Fenix CI startup tests have failed due to a seccomp panic after a `recvmmsg` call.
-
Upgrade mozilla-central to quinn-udp v0.5.6 ready for review.
-
Upgrade was previously blocked due to version conflict in transient depedency, namely windows-sys v0.52.
-
After larger discussion, we found a solution not involving upgrading many crates in the Rust ecosystem.
-
-
-
-
milestone 3 - Benchmark result for download and upload
-
Optimize neqo-client send path for more accurate benchmark results.
-
Simulating gbit link up- and download via Neqo network simulator.
-
-
Lars' quinn-udp pull request merged, adding experimental multi-message support on Apple platforms.
-
Adjusted quinn-udp benchmark to measure GSO, GRO and recvmmsg performance.
-
-
milestone 5 - multi-packet state machines
-
Pull request stack, removing major allocations in Neqo UDP receive path is merged.
-
Mozilla-central upgrading to latest release (v0.10.0) containing the optimization happening in phabricator#D226823.
-
Minor follow-up for the Neqo test server is pending.
-
-
Identify where bottlenecks are is it GSO or other
-
-
Implement WebTransport based on HTTP/2
- – enter updates –
-
Performance work:
-
[acreskey]
-
Expecting to start these four experiments tomorrow
-
https://mozilla-hub.atlassian.net/wiki/spaces/FIREFOX/pages/11047744/Experiments+Necko#Schedule
-
https://experimenter.services.mozilla.com/nimbus/fetch-priority-fenix-nightly/summary (Fenix nightly)
-
https://experimenter.services.mozilla.com/nimbus/fetchpriority-beta-pageload-event/summary (Desktop beta)
-
https://experimenter.services.mozilla.com/nimbus/performance-metric-calibration-with-lcp-nightly/summary (A/A/A/A on Desktop nightly)
-
https://experimenter.services.mozilla.com/nimbus/fenix-performance-metric-calibration-beta-followup/summary (A/A/A on Fenix beta)
-
-
-
Based on Randell’s profiling of main thread, ran this test and logged this one:
-
Great help from Sean Kim:
-
Network Marker Profiler upgrades
-
-
[sunil]
-
OnDataFinished and OnStopRequest probes indicate we still experience delays in between dispatching and running OnDataFinished in the background thread.
-
Did local browser tests to confirm this. ChannelEventQueue design seems to be a possible bottleneck. Will investigate further this week.
-
-