nuwa - process size reduction on b2g

Lloyd Hilaiel lhilaiel at mozilla.com
Wed Nov 13 10:08:35 PST 2013


On Nov 13, 2013, at 7:19 PM, Fabrice Desré <fabrice at mozilla.com> wrote:

> Hey Lloyd,
> 
> On 11/13/2013 08:23 AM, Lloyd Hilaiel wrote:
>> Disclaimer: this is all hopeful future rumination, and changes nothing w.r.t to our initial strategy for landing FxA on FirefoxOS in the 1.3 timeframe. I feel like this is important work with important implications, so I’m detailing it here.
>> 
>> So I was able to see this magic, and it was inspiring.  Basically when Nuwa lands (target looks to be 1.3) we’re going to see 3x faster app startup and 66% memory usage.  This means a little tiny bare Firefox Accounts process would live in 6.3mb of USS and spawn in half a second.
> 
> I don't know how you measured your startup time, but the 1900ms you
> mention in the bug seems pretty bogus to me.

Aha!  Thank you for highlighting an anomaly that I’d really love to understand.  I’m *am* measuring more than just application startup.  I’m messaging the time it takes to send a message over IAC to an application and to receive a response.  Here’s the tree that I’m using as a benchmark (a pile of commits you can plonk on latest gaia if it makes you giggle):

https://github.com/lloyd/gaia/compare/72a2b3a...master

What I see is after a reboot, launching the test firefox accounts application, and then clicking the button which triggers IAC and round-trips a message, takes 1900ms on a ZTE open.

If you kill the Firefox Accounts process and perform the same thing again it takes in the neighborhood of 500ms to re-spawn and complete the messaging round trip.

If you hit the button while the process is running it takes about 50ms to get there and back

If you membust causing all apps on the phone to be shutdown, and restart test firefox accounts, you are back to the 1900ms number.

I cannot explain these phenomenon.

> The template app on
> datazilla
> (https://datazilla.mozilla.org/b2g/?branch=master&device=hamachi&range=7&test=cold_load_time&app_list=templateo&app=template&gaia_rev=df92f0ebd89efbc6&gecko_rev=d62ac3ed7a650878)
> currently starts in about ~200ms.

there’s a big difference between 1900ms and 200ms.  There’s still a big difference between 200ms and 500ms.  Interesting.  Is it possible I’m talking worst case, and you’re talking average or best?

For marketing purposes, we should obviously use the case that gives us the most powerful numbers and then say “up to 3x faster”.  I’ve heard that works well ;).

lloyd

> I'd love to see a 3x speedup but I
> won't hold my breath on it ;) We have a potential improvement coming in
> that may save a few tens of ms but that has nothing to do with nuwa
> processes.


> The memory savings are in line with the expectations though.
> 
> 	Fabrice
> -- 
> Fabrice Desré
> b2g team
> Mozilla Corporation




More information about the Dev-fxacct mailing list