Closed
Bug 923949
Opened 11 years ago
Closed 10 years ago
[Template] Template application is slow
Categories
(Firefox OS Graveyard :: Gaia, defect)
Tracking
(Not tracked)
RESOLVED
FIXED
1.3 C3/1.4 S3(31jan)
People
(Reporter: kgrandon, Assigned: kgrandon)
References
Details
(Keywords: perf, Whiteboard: [c=progress p= s=2014.01.31 u=])
The current template app startup time is ~300ms. I think the goal should be < 100ms. Improving this will improve all apps.
Assignee | ||
Comment 1•11 years ago
|
||
Some notes: It takes between 20ms and 40ms from homescreen tap event handler until the system app gets the webapps-launch mozChromeEvent. From the time that the system app gets the message and actually launches the app, it takes ~100 ms. I think the first step is to see if we can lower the amount of time that the system app is taking here. Also had a chat with Vivien and it appears that we are doing some work to move CSP into C++. This will mainly help apps which load tons of resources, but it may help the template app a bit also.
Assignee | ||
Comment 2•11 years ago
|
||
Here is a profile of template app startup: http://people.mozilla.org/~bgirard/cleopatra/#report=d0ca0a82c09db05f0f71784907dfa62a391ac0fb
Comment 3•11 years ago
|
||
There is about 140ms of painting at time 2660. That seems somewhat excessive given there is almost zero DOM and CSS. It appears to be context switching between the parent and child process during this period. Is this expected? What IPC is necessary during painting?
Assignee | ||
Comment 4•11 years ago
|
||
Hey Matt, Benoit - I was told that you guys might be a good resource to look into this profile. If you have the time, I would greatly appreciate if you could take a quick look. Thanks!
Flags: needinfo?(matt.woodrow)
Flags: needinfo?(bgirard)
Comment 5•11 years ago
|
||
Any chance we can throw in a marker in the b2g process from the point where it decides to start the other process? It's not clear at which point the main process decides to start the other app before which we're not interested in the data.
Flags: needinfo?(bgirard)
Comment 6•11 years ago
|
||
The time spent during |IPDL::PBrowser::RecvShow| is a bit concerning. It looks like it delayed by the main process handling other queued events before it can get to them. Particularly we have two long refresh ticks. Perhaps its the app open animation but we need to dig down and figure out what is being painted during that time. There's also |IPDL::PContent::RecvAsyncMessage| which I'm not familiar with which is concerning.
Comment 7•11 years ago
|
||
Kevin, if this is really a blocker to bug 920713 please koi? it as well as its blocker bug 924032.
Flags: needinfo?(kgrandon)
Whiteboard: [c= p=4 s= u=] → [c=progress p=4 s= u=]
Assignee | ||
Comment 8•11 years ago
|
||
Noming as this blocks a blocker.
blocking-b2g: --- → koi?
Flags: needinfo?(kgrandon)
Assignee | ||
Comment 9•11 years ago
|
||
Going to hold off on koi nomination for now until we can ensure that this is actually a regression. Bug 920713 uses a different Hello World application, so we could see if we can standardize on a single app. Unblocking for now.
No longer blocks: 920713
blocking-b2g: koi? → ---
Updated•11 years ago
|
Flags: needinfo?(matt.woodrow)
Assignee | ||
Comment 10•10 years ago
|
||
Nothing immediately actionable besides investigation, and there has been a bunch of improvements here recently. Closing this.
Status: ASSIGNED → RESOLVED
Closed: 10 years ago
Resolution: --- → FIXED
Whiteboard: [c=progress p=4 s= u=] → [c=progress p= s=2014.01.31 u=]
Updated•10 years ago
|
Target Milestone: --- → 1.3 C3/1.4 S3(31jan)
You need to log in
before you can comment on or make changes to this bug.
Description
•