If you think a bug might affect users in the 57 release, please set the correct tracking and status flags for Release Management.

Scrolling on some specific pages causes Firefox to use HDD constantly and loudly

NEW
Unassigned

Status

()

Toolkit
Places
P5
major
3 years ago
2 months ago

People

(Reporter: Virtual, Unassigned)

Tracking

(Blocks: 1 bug, 4 keywords)

Firefox Tracking Flags

(Not tracked)

Details

(Whiteboard: [qf:p3], URL)

Attachments

(1 attachment)

1. Open this URL website page - http://pubchem.ncbi.nlm.nih.gov/compound/656612?from=summary#section=Top
2. Scroll page to the bottom and next to the top a few times
3. You can clearly hear that your HDD is used loudly and constantly by Firefox



The issue doesn't occur in in:
-Chrome (Blink engine),
-Opera (Blink engine),
-Opera (Presto engine) [poor scrolling performance];
but Vivaldi (Blink engine) is also affected.

Comment 1

3 years ago
Created attachment 8571085 [details]
hash.html  possible testcase.

possible testcase.

Updated

3 years ago
Component: Untriaged → Places
Product: Firefox → Toolkit
I see lots of disk usage. But it doesn't majorly affect scroll speed, so how is this topperf?

Comment 3

3 years ago
Regression window with testcase of comment#1
Pushlog:
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=e435c9812855&tochange=cd62a8afc52e

Comment 4

3 years ago
Also occurs on http://rr-project.org/rr.html#1.0 .  As soon as a page is scrolled, Nightly writes to the disk. Quickly scrolling the pages will peg the HDD constantly
Flags: needinfo?(mak77)
the testcase is not very common, so not a priority, but still worth investigating.
Unfortunately the regression range is not really useful, too old and widespread.
Chrome is storing history for each fragment, as we do, in a similar table, so the disk load should be ideally similar.

First thing to notice from the log, we seem to read the page info twice (double calls to FetchPageInfo), likely one can be avoided (to be verified).
Then we fetch the visit id, the only reason is to notify onVisit, but it's unlikely a useful information, we we could stop notifying that and save a read.
Then we update Frecency for the url, this is expensive, we could evaluate some way to delay it so that we do only if a further visit to the same page is not seen in a timeframe (so we don't update too often).

So, there is definitely some space for improvements, but globally I think most of the churn here comes from keeping frecency updated.
Blocks: 691507
Flags: needinfo?(mak77)
Keywords: topperf

Updated

2 years ago
Depends on: 1209027
Has STR: --- → yes
Keywords: nightly-community

Updated

6 months ago
Priority: -- → P5
Whiteboard: [qf:p1]
Whiteboard: [qf:p1] → [qf]
Whiteboard: [qf] → [qf:investigate:p1]
Maybe my profile will has some hint and speed up diagnosing & investigating this issue.

Profile: https://perfht.ml/2opHq6q
Taken with: Gecko Profiler 2.1.0
Interval: 0,01 ms
Buffer size: 900 MB
Threads: ,
Features: Stack walk + JavaScript + Task tracer

Updated

4 months ago
Whiteboard: [qf:investigate:p1] → [QRC][QRC_NeedAnalysis][qf:investigate:p1]
Keywords: regression
Whiteboard: [QRC][QRC_NeedAnalysis][qf:investigate:p1] → [qf:p3]
QA Contact: Virtual
You need to log in before you can comment on or make changes to this bug.