462280 - profile Tb startup time with and without Lightning

Status: RESOLVED WONTFIX

(Calendar :: Lightning Only, defect)

Description

[Dan Mosedale (:dmosedale, :dmose)]

Since this seems likely to cause a startup-time hit, we need to understand what that looks like given some reasonable set of calendar and email data before we turn it out in the nightly builds.  This may depend on Ts work that Standard8 has pending, so getting a status update there seems like a good start.

Comment 1

[Mark Banner (:standard8)]

Now we have the bloat tests running, I'll be able to look at this a bit more.

The two critical parts to this work will be:

1) deciding on what data to include in the profile.
2) working out where to hook into the app to determine a reasonable Ts.

I think some work has already been done for Sunbird on 2. I'm currently trying to contact folks to find it, and will most likely leverage that. This should be fairly simple to write now we have the bloat test harness.

The current version of the bloat tests just has some accounts set up and doesn't load any data. I'm willing to expand that and use the same profile for Ts. See https://wiki.mozilla.org/MailNews:Memory_Bloat_and_Leak_Tests for what it currently does have set up.

I think probably the best thing here is for me to come up with some ideas for a profile from the MailNews side, and someone to come up with some ideas for Lightning, and we'll put them together in a proposal and post to the newsgroups.

Comment 2

[Dan Mosedale (:dmosedale, :dmose)]

Marking as P1, as I think having an order-of-magnitude understanding of what sort of hit this is likely to cause seems worth having sooner rather than later.

Comment 3

[Pete Riley]

If you want to see how Lightning causes Thunderbird to start very slowly, I suggest that you subscribe to some calendars with a lot of recurring events, such as holiday calendars.

Comment 4

[Miles]

Here's my take on the speed for my old computer with WinXP Home, SP2 w/TB 2.0.0.17.  Start up time in seconds:
1.  Safe mode = 5 to 7
2.  With all extensions enabled including Lightning = 45 but after several      restarts reduced to 40
3.  With only Lightning disabled = 15
4.  With all disabled except Lightning = 30
5.  All disabled including Lightning = 5 to 6
6.  Enabled Lightning only = 30 
7.  Enable all extensions = 40 

The speeds were tested several times with and without reboot.
Number of extensions = 34 including Lightning + 5 disabled.
Number of calendars = 4 Holiday, Mine, and Moon Phases = 6.
Theme SkyPilot Classic was left on throughout (except safe mode)

Apparently Lightning doubles TB's startup time and considering I start TB several times a day, it is quite noticeable.  Due to the foregoing would prefer to use Sunbird, but haven't found a method to auto receive reminders unless it is running, as opposed to Lightning where notices are received when TB is running, even if focus is not on Lightning.

Comment 5

[Mark Banner (:standard8)]

Miles, thanks for the interesting data, it will certainly be interesting to compare that with the current trunk builds (there are a lot of gecko performance enhancements that should help us and may hide some of this).

While I'm here, a quick update on getting a test harness: I've currently got the beginnings of an extension (based on some work by mvl) that I think will be able to measure Ts at a consistent point. I'm currently trying to resolve a few issues, but hopefully will having something running early next week at least locally and possibly runnable as an extension plus a script.

Comment 6

[Miles]

Neglected to mention that I'm using Lightning 0.9.   Is there a "quite stable" beta upgrade that I could change to in order to see if start up time is reduced?

Comment 7

[Mark Banner (:standard8)]

In bug 462862 I have now posted a test extension that implements a harness that should allow reasonably consistent measuring of start up time for Thunderbird.

I have not tried it with Lightning installed, and it probably needs some tweaks before the final version, but it may be good enough to start giving us figures here.

I think we should still attempt to define a standard profile to test against. Though at this stage I guess rough details of profiles will be sufficient to help us get some rough ideas on the time with and without Lightning.

Thoughts?

Comment 8

[Peter Weilbacher]

Unless i misunderstand the problem, I think it would also be interesting to know how large the calendars are that Lightning has to load. I also notice that TB with Lightning is quite slow to start up (just like Sunbird), when I have it configured to load my normal 600 kB ICS calendar. (And network access covers only 10% of that time.)

Comment 9

[Mark Banner (:standard8)]

If anyone wants to take this on, Bug 462862 now has an extension and python script that can start up with Thunderbird and measure the time (it was used for bug 427627).

Obviously we'd probably want to check with no calendars and reasonable size calendars (and maybe network ones), but just no calendars would be a start.