Closed
Bug 1960583
Opened 1 year ago
Closed 1 year ago
Revert minimum pacer delay constraint
Revert minimum pacer delay constraint
| Tracking | Status | |
|---|---|---|
| firefox139 | --- | fixed |
---
---
[necko-triaged]
QA Whiteboard:
[qa-triage-done-c140/b139]
Has STR:
---
Change Request:
---
0
Bug Flags:
| Assignee | |
Description•1 year ago
|
https://phabricator.services.mozilla.com/D191325 introduced an optimization, practically speaking ignoring any pacer delay < 1ms.
Since then https://github.com/mozilla/neqo/pull/2035 fixed the underlying issue in Neqo itself.
I suggest removing the optimization in neqo_glue, see also past discussion, for the following reasons:
- reducing code duplication
- instead of messing with time, always provide accurate time (i.e.
Instant::now) to Neqo
Pushed by mail@max-inden.de:
https://hg.mozilla.org/integration/autoland/rev/36d26905ef58
Revert minimum pacer delay constraint r=kershaw,leggert,necko-reviewers
Comment 3•1 year ago
|
These prefs are removed:
network.http.http3.max_accumlated_time_ms
Blocks: old-prefs
Comment 4•1 year ago
|
|
| bugherder | |
Status: ASSIGNED → RESOLVED
Closed: 1 year ago
status-firefox139:
--- → fixed
Resolution: --- → FIXED
Target Milestone: --- → 139 Branch
Updated•1 year ago
|
QA Whiteboard: [qa-triage-done-c140/b139]
Posting the results of this patch here in order to be able to link to a public resource.
This patch improved 95th percentile and 75th percentile of http3_timer_delayed metric. The 75th percentile is now below or close to our 1 MS timer accuracy limit.
Improved timer accuracy improves various QUIC components, e.g. RTT measurement and thus congestion control.
You need to log in
before you can comment on or make changes to this bug.
