Closed Bug 1173575 Opened 11 years ago Closed 4 years ago

www.hatena.ne.jp takes 2x as long as Chrome to load

Categories

(Core :: General, defect)

defect
Not set
normal

Tracking

()

RESOLVED WORKSFORME

People

(Reporter: rbarker, Unassigned)

Details

Attachments

(2 files)

While loading pages such as http://www.hatena.ne.jp the downloading of resources can stall for over a second at a time. Reproducible on both desktop and android. The same pages loaded into chrome do not show the same delays.
The two attached images show the same page loaded in Fennec(22 sec) and Chrome for Android(11 sec) running on a Nexus 4. Fennec has a number of these stalls while chrome does not and completes the page load much sooner than fennec. Similar results can be seen on desktop.
I'm gonna bet this has to do with an interaction with the main thread and javascript blocking it. We had talked about nate taking a look at this origin already, I think its time to do it.
Flags: needinfo?(jduell.mcbugs)
So I'm trying to understand what "stall" means here. From a glance at the screenshots, it doesn't look like channel start-end load times (i.e. the time taken between AsyncOpen and OnStopRequest) are noticeably different than Chrome's (it's hard to tell, since the screenshots show a disjoint set of subresource URIs). Randall, do you have a sense how much the overall page load difference here is the result of it taking us longer to issue channel loads (vs how long the actual loads take once they are issued)? If the problem is that we're slower than Chrome at starting subresource loads, but the loads themselves take similar time, then this most likely isn't a necko issue, but something higher up taking too long to get around to asking for URI loads. A good start here might be to find out what code asks for the analytics.js URI, since that's taking forever to get started. But of course it could be just about anything clogging up the main thread. Maybe time to run the Gecko profiler?
Flags: needinfo?(jduell.mcbugs)
Note that we've run into this sort of thing before--see bug 1062713 comment 10 through 25. We could similarly instrument a patch for hatena.ne.jp, but it looks like we already have a visualization of that from the devtools screenshots.
Flags: needinfo?(rbarker)
The difference in load times for my device (nexus 4) is 22 seconds for fennec and 11 seconds for Chrome for Android. It was hard to show in the screen shots since the page has a lot of resources however in chrome, there are no gaps in the start of resource downloads and the timeline is smooth while in fennec, the time line has huge gaps between resource downloads as I tried to show in the screen capture. So, if you look at the chrome time line it looks like what is shown in the screen shot over the entire time line. The fennec time line has several gaps which cause the page load to almost double in time to load vs. chrome.
Flags: needinfo?(rbarker)
Right, so the issue is that we're somehow delaying the start of network loads, which is usually symptomatic of something above necko getting bogged down. Someone needs to run the Gecko profiler and figure out what that is.
Component: Networking → General
Summary: Network stalls while downloading resources → www.hatena.ne.jp takes 2x as long as Chrome to load

On Windows 10 desktop, I tested Chrome Version 96.0.4664.110 (Official Build) (64-bit) against Firefox 95.0.2 and got essentially same response ~3 seconds

Status: NEW → RESOLVED
Closed: 4 years ago
Resolution: --- → WORKSFORME
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: